Fastmail Release Notes

Follow

15 release notes curated from 1 source by the Releasebot Team. Last updated: Oct 29, 2026

Get this feed:
  • Oct 29, 2026
    • Date parsed from source:
      Oct 29, 2026
    • First seen by Releasebot:
      Oct 29, 2026
    Fastmail logo

    Fastmail

    Keep your home server reachable with Dynamic DNS

    Fastmail adds Dynamic DNS support for hosted domains, keeping home-hosted services reachable as residential IP addresses change. Customers can update opted-in DNS records through the dyndns2 protocol or a simpler API, with scoped tokens that protect other DNS records.

    This post is about a new feature for people who host their domain’s DNS with Fastmail, and it’s a technical one. You don’t need to change anything about your account or your email software, so if running your own services at home isn’t your thing, feel free to skip it.

    Today we’re announcing Dynamic DNS support in Fastmail. It speaks dyndns2 at https://api.fastmail.com/dynamicdns/update, and there’s a simpler one-line variant for a self-hoster’s crontab.

    The problem with a home IP address

    Some customers, myself included, self-host some services at their homes. Residential ISPs typically don’t provide a static IP address. It might change daily, monthly or just often enough to be a pain. The problem this creates is you set up home.various.domains pointing at your home IP address, but eventually your IP changes and the hostname doesn’t point to you. Your NAS is still running and your home automation is still ticking, but now you can’t reach them. dyndns2 is a protocol that has solved this problem for 25 years and is supported by many home routers and popular clients such as ddclient.

    Getting started

    To get started, create a new API token of the Dynamic DNS type in Settings → Privacy & Security → Manage API tokens. Copy the token somewhere safe before you navigate away as it’s only shown once. Then go to Settings → Domains, click Edit next to your domain, then Customise DNS, and add a record of type Dynamic DNS with the name you want to keep up to date. The record starts out empty and is set via the API.

    The security of the Dynamic DNS type

    We built this as a new pseudo record type (an internal, non-standard type that resolves to an A or AAAA record), in the same vein as an ANAME record. Its value can only be set through the API, never in the web interface, so creating one is an explicit opt-in. That is what limits a Dynamic DNS token: it can only write to the records you’ve opted in, and can’t touch an A or AAAA record, your MX records or your DKIM keys. Therefore a token sitting in plain text in your basement can’t break your mail.

    The dyndns2 API

    If you have a dynamic DNS client on your home server, or your router supports it, you’re ready to go. Point it at https://api.fastmail.com/dynamicdns/update. dyndns2 uses HTTP Basic auth, with your Fastmail login as the username and the API token as the password. For compatibility with some older clients we also alias /nic/update to /dynamicdns/update.

    Here’s a working config for ddclient 3.10.

    # /etc/ddclient.conf
    ssl=yes
    protocol=dyndns2
    server=api.fastmail.com
    [email protected]
    password=********
    home.various.domains
    

    Substitute your own username next to login and replace the ******** with your API token, not your account password. ssl=yes is required, as we don’t want anyone sending credentials over plain text. Keep it owned by root and mode 600, since the token sits in it in plaintext.

    The simple API

    We added a simpler API too, for the selfish reason that some of us (me) live and die by our crontab. It’s perfect for a one-line curl entry.

    */15 * * * * curl -fsS "https://api.fastmail.com/dynamicdns/simple/home.various.domains/********/"
    

    There’s no username required; the API token tells us which user it belongs to. The -f makes curl fail on HTTP errors and -S lets that error through, so cron only mails you when an update breaks. For convenience, omitting an IP address will set it to the request IP. Or set it explicitly, with IPv4 /dynamicdns/simple/home.various.domains//192.0.2.1 or IPv6 /dynamicdns/simple/home.various.domains//2001:db8::1.

    The response body is always empty and the result is carried in the status code.

    • 204 means the record is set, whether or not it changed
    • 400 means we couldn’t parse the hostname or the address
    • 401 means the token is wrong
    • 404 means that name has no Dynamic DNS record

    Why this matters if you host your DNS with us

    Dynamic DNS isn’t going to improve your email, so why did we add this feature? DNS is vital for email, and these days it isn’t only MX records. It’s making sure SPF and DKIM are set up correctly too, and that list is only going to get longer. For some customers, not having Dynamic DNS support was a blocker to hosting their DNS with us. Remove that blocker and we can take care of all that complexity for you.

    Dynamic DNS is available now on any domain you host with us, and our help page covers the full API.

    Not a Fastmail user yet? Discover all our powerful features to make you more productive, or start your 30-day free trial today.

    Original source
  • Oct 15, 2026
    • Date parsed from source:
      Oct 15, 2026
    • First seen by Releasebot:
      Oct 15, 2026
    Fastmail logo

    Fastmail

    A calendar for every time zone, and a lot more you can do with your events

    Fastmail releases a major calendar update with world clocks and time-zone views, attendee local times, easier event and calendar actions, and a cleaner, faster experience. Duplicate events across calendars now appear just once, helping users manage busy schedules with less friction.

    Today we’re releasing a big update to Fastmail’s calendar. If you work with people in other time zones, you can now keep their time on screen as you plan your day. Every event has a full set of actions a right-click away, and events that appear on more than one of your calendars are finally shown just once.

    A world clock for your calendar

    Head to Settings → Calendar preferences and you’ll find a new World clock section. Add the places you care about, whether that’s where your team is, where your family lives, or where you’re travelling next.

    Once you’ve added some, the current time in each one appears in the calendar sidebar on desktop, under the mini calendar. Click any of them to see your whole calendar in that time zone. It’s handy for working out what your day will look like when you land.

    Extra time zones on day and week view

    In day and week view, each of your world clock time zones can get its own column of hours alongside yours, so you can see at a glance that 9am for you is 5pm in London. The current time line runs across them all too.

    Choose how much room they take up: show them always (if there’s space), only on hover or tap, so they slide out when you point at the hours, or never. You can change this in the world clock settings, or right from the calendar.

    Click the name at the top of an extra time column for a menu: view your calendar in that time zone, remove it, add another time zone, or jump to the world clock settings. You can also drag the columns around to reorder them. Drag one onto the main column to switch your calendar to that time zone.

    When you’re viewing your calendar in another time zone, your own time zone now always stays on screen right next to it, so you never lose track of what time it is for you.

    Know when it is for everyone invited

    When you open or edit an event with other participants in your organisation, we’ll now show what time it is for them too, if they’re in a different time zone to you. A 4pm meeting in Melbourne is a 6am one in London, and it’s good to know before you send that invitation. This works for all the other people in your Fastmail account.

    Do more with your events

    Right-click any event in the calendar, or long-press it on a touchscreen, for a menu of things you can do with it. You’ll find the same options in the More menu when you open an event.

    • Duplicate: start a new event with all the same details, ready to adjust and save. If it’s a meeting you organised, the same people are invited again.
    • Go to next/previous occurrence: hop through a repeating event without paging through weeks of calendar.
    • Email all attendees: start an email to everyone invited, sent from the address you were invited at.
    • Download: save the event as an .ics file to share or import elsewhere.
    • Edit and Delete, of course.

    The calendars in your sidebar have a menu too. Right-click one to quickly create an event in it, change its colour or name, or open its settings. Choose Show only this calendar to hide everything else for a moment, and Show all calendars to bring them all back. You can also drag and drop to reorder them, right in the sidebar.

    Clicking an event’s location now gives you options to open it in a map, get directions there, or copy it.

    One event, however many calendars it’s on

    If someone shares their calendar with you and also invites you to their meetings, you used to see every one of those meetings twice: once on their calendar and once on yours. Now, when the same event is on more than one of your calendars, we show it just once, striped in each calendar’s colour.

    When you edit or delete an event like this, we’ll ask which calendar you mean. And if it’s a meeting you organised, we’ll check whether you want to change the event for everyone, or just your own alerts and settings.

    A fresh look for events

    We’ve refined the event view to make the important details easier to see, with the calendar right at the top and the time before the location. Free/busy now only appears where it makes a difference, and when you edit an invitation from someone else, we only show the things you can actually change.

    Invitations you haven’t replied to yet now have a dashed outline, so they stand out from the events you’ve accepted. Ones you’ve declined are faded back (if you’ve chosen to keep them visible) so they don’t get in your way.

    And it’s faster

    Finally, we’ve made the calendar quicker when there’s a lot going on. Scrolling through busy months, and loading events as you go, now does less work behind the scenes, so it feels much smoother, especially with lots of calendars or repeating events.

    We hope these changes make it easier to stay on top of your schedule, wherever the people in it happen to be. Give them a try and let us know what you think!

    Original source
  • All of your release notes in one feed

    Join Releasebot and get updates from Fastmail and hundreds of other software products.

    Create account
  • Oct 8, 2026
    • Date parsed from source:
      Oct 8, 2026
    • First seen by Releasebot:
      Oct 15, 2026
    Fastmail logo

    Fastmail

    Group your mail: keep the important stuff together, right at the top

    Fastmail launches customizable inbox groups that organize messages by age, pinned status, unread status, or search criteria. Available across plans, the web, and apps, groups sync across folders and saved searches while keeping existing sorting and notifications unchanged.

    Today we’re launching the ability to split your inbox into sections so the messages that matter most are always at the top, and related mail sits together. You can set groups in other folders or labels too, and even saved searches.

    Grouping is available on every plan, on the web and in all our apps. It’s entirely opt-in: nothing changes until you turn it on.

    Why groups?

    Most inboxes are sorted one way: newest first. That’s simple, but it treats every message the same. The email from your team, the calendar invite waiting for a reply, and the “your parcel has shipped” notification are jumbled together in one long list.

    With our pinned-on-top feature, you could manually push important messages to the top, but mailbox groups take this a step further by letting you pre-sort your inbox automatically.

    Ready-made groupings

    Open the More menu (⋯) at the top of your message list and choose a grouping:

    • By age: today, yesterday, this week, this month, and older. Perfect for catching up after a few days away.
    • Pinned first: the conversations you’ve pinned, then everything else.
    • Unread first: anything you haven’t read yet at the top, with the rest below.

    Each section header shows how many conversations it holds. Click a header to collapse a section you don’t need right now (we’ll remember that for next time), or use Select all to archive or move a whole group at once.

    Grouping doesn’t replace sorting. Choose whatever sort order you like, not just newest first, and it applies within each group.

    Make your own

    This is where it gets really powerful.

    Choose Custom… to build exactly the groups you want. Start from a few suggestions, like Pinned, Unread, Following, Calendar invites, or VIPs. If you use your own domain, there’s also an option to group mail from everyone at your organisation. Or write your own, using the same search syntax you already know, for example:

    • from:@github.com — keep high-volume notifications separate
    • list:jmap.ietf.org — your favourite mailing list
    • to:[email protected] — anything sent to a shared alias
    • subject:invoice OR subject:receipt — keep an eye on your purchases

    Name each group, and drag and drop to reorder. If a conversation matches more than one group, it appears in the first one. Anything that doesn’t match ends up in a final group, which you can name whatever you like.

    Groups are set separately for each folder, label, or saved search, so your inbox can be split into work, notifications and everything else, while your Archive is grouped by age. They’re saved to your account and sync everywhere.

    See it in action

    A few things to know

    Groups work using search operators like from:, to:, subject: and is:. They can’t search the body of a message, so plain words like meeting won’t work as a group on their own. If you try something that isn’t supported, we’ll tell you which part to change.

    Grouping works whether or not you have conversations turned on. With conversations off, each message is placed in a group individually.

    Groups only change how your message list is laid out. Unread counts and notifications work exactly as they did before.

    We hope groups help you get on top of your inbox. Give them a try and let us know what you think!

    Original source
  • Aug 19, 2026
    • Date parsed from source:
      Aug 19, 2026
    • First seen by Releasebot:
      Oct 15, 2026
    Fastmail logo

    Fastmail

    Data resilience at Fastmail

    Fastmail explains how its backups restored mailboxes affected by a Cyrus replication bug, preventing email loss, and describes a fix with a test to catch the issue in future.

    (or: How I Learned to Stop Worrying and Love the Backups)

    This is a story of a close call, where users lost entire mailboxes due to a bug, but backups allowed us to bring everything back.

    We’ve been doing this quite a while

    This is a blog post about principles. Principles rarely change, which is why you can read a blog post I wrote over 10 years ago and the “three types of data” is still as fresh as the day I wrote it.

    This is also a blog post about specifics of implementation. We don’t change that very fast either, which is why almost everything else in that post is also still correct! We’re still using git, we’re still using Debian. Some mechanics about the Makefiles and search database compaction have changed, but we still do them too.

    And importantly, we still store data replicated across multiple machines, but today that guarantee is even stronger. At least one of those machines is in a different physical location (currently Philadelphia, St Louis, or Amsterdam). The default configuration is two machines in your primary location plus a third machine offsite. We have sufficient compute resources to continue operating even if our biggest site was temporarily offline.

    Here’s another thing we’ve been running for a long time — an in-house backup system for Cyrus IMAP, the server where most of our customer data is stored. We blogged in 2015 about a replacement system, which was eventually set aside and we kept improving the in-house system instead. That system is now available as part of the open-source Cyrus project. It will feature in this story.

    A nasty bug

    In April 2026, Fastmail decided to segment accounts into separate “stores” (individual Cyrus instances) for EU and non-EU customers, in advance of setting up our Amsterdam location.

    As part of this, most EU customers were migrated to newly created stores, using the existing Cyrus-replication based method for moving users. This mechanism is regularly used to balance storage with automatic moves, as well as to move users together when accounts are combined — however in this case, a much higher volume of moves were performed over the course of about a month.

    During this time, 8 user accounts had a subset of their folders entirely wiped by a bug. In every case that included their INBOX, rendering the entire account inaccessible. As is usual with software bugs, it was caused by the interactions of multiple behaviours which seemed sensible in isolation, but interacted badly.

    For those interested in the full technical details of what happened, read on. Otherwise feel free to skip ahead to the next section to see how backups ensured no data was lost.

    The proximate cause was a well-intentioned change to the no-copyback replication mode, to forcefully clear out any remote state for the user being synced. This was in pull request 5513. We only use no-copyback mode at Fastmail when we’re moving a user to a new store. There isn’t going to be anything there for the user yet, so being more aggressive makes sense.

    The bug worked in conjunction with another change from pull request 4852, which ensured that when syncing a user, we follow its rename history.

    Cyrus keeps tombstones with old names for every mailbox’s uniqueid for some time, to allow renames to be detected during regular replication, and to ensure we never re-use a UIDVALIDITY if the invariants that IMAP requires cannot be satisfied.

    Individually, these all make complete sense. Combine them and there is a big problem — reused usernames. We made a decision way back in the early days to use the primary email address directly as the username in Cyrus, so folder sharing looked right via IMAP. This has caused some pain over the years! We reserve “used” usernames on our own domains for a while, but if you’re creating accounts in your own domain, then immediate re-use becomes possible.

    Consider this case: user A has been renamed to B, and then a new different user A is created — same username, same store. We sync user A to a different server. This happens fine, the integrity checks pass, all good so far.

    Later, we try to sync user B. The name history of B’s folders includes folder names for user A. The sync for user B reaches into the “new” user A on the destination server and wipes any folders with those names. For these 8 accounts that’s exactly what happened — the move system wiped entire folders on all three of the target store’s copies!

    The initial attempt at a fix was in pull request 5961, which didn’t actually stop the issue. It was based on an incorrect assumption that the user was coming from a separate store, and the tombstones were on the destination rather than the other way around. Due to the incompleteness of the fix, one user who had not been fully moved before the initial restore from backup was corrupted again, and this was only finally fixed with pull request 5988.

    This only happened for Fastmail users who had created a user, renamed it to something else, then re-created the same original username. And of those, only users who were moved during the period when the bug existed.

    Backups save the day

    Fastmail runs cyrus-imapd with a slots and stores layout as documented on our help pages. The replication protocol has multiple safeguards built in, checksums on everything, and regular rolling replication can never wipe something immediately, only mark it for later cleanup in a week or so.

    This protects against damage by regular users, and most possible bugs. This particular bug cut straight through those protections though, by using the replication protocol to immediately wipe the folders. The move system needs to be able to delete things, because as its final step after ensuring that a user has been successfully migrated, it cleans up the copy on the source store!

    Fastmail also has a backup system, which stores all the data and metadata in an append-only incremental backup with occasional repacking, meaning that all data from the most recent backup was safe. We regularly back up each user every few hours, and also immediately before and after moving their account between servers, which is what saved us here!

    All those missing folders were restored from backup, recovering all emails for all but one user.

    For that one user, 5 sent emails were unrecoverable from backups. Luckily, identical copies of those sent emails were recoverable from the #jmapsubmission temporary files inside Cyrus, using some additional custom-built tooling. Ultimately, no email was lost.

    Future protection

    The solution in 5988 includes a test case which reproduces the exact pattern that was actually happening during those moves, which means we will detect if we write a similar bug again.

    We’re also considering moving the replication protocol to be more based on the user’s inbox uniqueid for everything rather than mboxname, which would have avoided this problem entirely, but that’s a bigger change that will likely be a couple of years away.

    Regardless of the ongoing improvements, the core principle which saved us here was exactly what I wrote back in 2014: “Primary Copy - back it up. Replicate it. Everything you can to ensure it is never lost. This stuff is gold”. By building defence in depth, we ensured that even this nasty bug didn’t lead to us losing the data we had taken responsibility for.

    Original source
  • Aug 3, 2026
    • Date parsed from source:
      Aug 3, 2026
    • First seen by Releasebot:
      Aug 5, 2026
    Fastmail logo

    Fastmail

    Fastmail offers EU data region

    Fastmail introduces EU data residency, letting customers choose Europe or the US as the primary home for their email and files. It now stores EU-region data on its own Amsterdam servers, with encrypted replicas and easy region switching in settings.

    US or EU? Your choice

    Many of you have been telling us that where your email is stored matters to you. There are various reasons for this, like local law, keeping your data close to home, or compliance.

    You can now make the European Union the primary home for your Fastmail data, on our own secure servers in Amsterdam. Previously, all accounts were stored entirely in the US. Now you have more choice.

    Built by us, not rented from someone else

    We’ve installed our own servers, co-located in a secure facility in Amsterdam, set up by our own engineers. This new location is built to the same high standards as our existing infrastructure in Philadelphia and St Louis, with our own hardware and our own software — specified right down to the exact model of disks in each machine.

    In all our locations, data is stored encrypted at rest inside locked racks, and managed by our in-house team. We don’t rent computing or management services from a big cloud provider and pass on their assurances. That’s how we’ve approached privacy, reliability, and performance for more than 25 years.

    For many years, we have kept at least two copies of everybody’s email on separate servers in their primary location, and at least one more in a geographically separate location to ensure data safety.

    Here’s the detail about where your data will flow based on your regional choice.

    If your account is in the EU region:

    Your primary copy of data will live in the EU. The main, live copy of your mail and files will sit on our own servers in Amsterdam.

    Incoming mail will go to EU servers by preference if you are using your own domain and Fastmail’s nameservers, or an address at one of Fastmail’s EU-region domains.

    Our desktop, web, and mobile apps will connect to EU servers. Day to day, our apps will talk directly to our Amsterdam infrastructure. If those servers are ever unavailable, connections fall back to one of our US locations so you can still reach your mail.

    Resilient replicas of your data will live in the US (for now). As we only have one location in Europe so far, the geographically separate copy will remain on servers in one of our US locations.

    If your account is in the US region:

    Most of your data lives in the US. Both the primary and replica locations of your mail and files will sit on our own servers in Philadelphia or St Louis.

    Incoming mail goes to US servers by preference if you are using your own domain and Fastmail’s nameservers, or an address at one of Fastmail’s US-region domains.

    Our desktop, web, and mobile apps will connect to US servers. Day to day, our apps will talk directly to whichever US location contains your primary data copy. If those servers are ever unavailable, connections fall back to our other US location so you can still reach your mail.

    What applies to everyone:

    We favour availability, so if your home location is down, you will temporarily connect to another of our locations so you can continue to access your data.

    If you are using one of Fastmail’s generic non-regional domains then your mail may go in or out via either location.

    Emergency backups for everybody are stored in our Philadelphia location. As well as the live replicas of your data, we also keep a separate set of encrypted backups taken every few hours for every account. These are in Philadelphia for all users at the moment.

    Some data is replicated to all sites, so parts of everyone’s data are in both Europe and America. This includes email addresses and other user/customer metadata, storage for websites and the standalone Files feature, and the details of any third party services you have linked.

    Logs are in the US. All system logs are consolidated into a single place for monitoring system health and to assist with customer support.

    Third party services are shared. The third parties we use for debugging, billing, and support are linked to your account in the same way regardless of your region.

    Your email client config (IMAP/POP3) does not have to change. The generic hostnames will proxy internally to your active server. We’re working on making those terminate at the closest network location. However, we also offer regional server names for each protocol to give you more control over where you connect.

    We’re an Australian company, subject to Australian law including legal-cooperation treaties between Australia and other countries. Wherever your data is stored, we will respond the same way to lawful requests from relevant authorities (see our transparency report).

    If what you need is a guarantee that your data remains only in the EU, we don’t have that, and we’d rather tell you directly than let you assume otherwise.

    We made a prediction, but you can change it

    We pre-selected all the users with billing addresses in or near Europe for the EU region. If you are one of these, an encrypted copy of your data was transferred to Europe in advance of this announcement, and will shortly become your primary copy. You can change region to the US and your data will migrate back.

    Conversely, if we didn’t identify you for our initial group, you can put yourself in the queue to be migrated. Moves this way will be a little slower because there’s no copy of the data already in the region, so we have to sync every email across the ocean! For those moving back, there’s already a copy in the US, so only a small amount of data needs to be synchronised to fully reconcile your mailbox.

    This blog post is all about giving you the facts and the tools to make the trade-off that’s right for you. Most providers decide for you and tell you as little as they can get away with. We’d rather show our work — what’s replicated, what isn’t, who can compel what — and let you choose with your eyes open.

    How it works

    When you sign up, you choose your region, and the primary copy of your data is placed on our secure servers in that region, with a copy also replicated to a geographically separate location. Regardless of your choice, all copies of your data are encrypted at rest. Choosing a region changes where your primary copy lives, not how well it’s protected.

    If you were with us when we selected the users to transfer, we’ve pre-set your region based on your billing address. If you signed up more recently, you’ll have been allocated to the US. Either way, if you’d prefer a different region, you can switch it in your settings.

    Switching region

    Go to Settings → Users & Sharing → Team Settings, then look below GDPR for the Data residency section. Pick your location and we’ll move your primary copy for you. You’ll see it go from queued, to transferring, to done, and your account keeps working the whole time. You can change your mind and move back, though we may place reasonable limits on how frequently you can change region!

    Your data, your call

    We’re really excited to have another location, and to be able to offer this option to you. Setting this up was a significant investment. We considered a surcharge for choosing the EU, but we don’t believe that’s right. We proudly charge a fair rate for an exceptional service, and this choice should be yours. Select the region that is right for you.

    Original source
  • Similar to Fastmail with recent updates:

  • May 25, 2026
    • Date parsed from source:
      May 25, 2026
    • First seen by Releasebot:
      Jun 16, 2026
    Fastmail logo

    Fastmail

    A support agent’s guide to getting your issue solved quickly

    Fastmail improves its support experience with a more user-friendly ticket form and optional AI Instant Answers, making it easier to submit detailed issues and get help faster. The post also highlights the value of clear ticket details and Fastmail’s responsive customer support.

    Hello! I’m Merlin! Some of you might recognize my name, as I’ve been working as a customer support agent at Fastmail for an impressive 11 years. It might sound like a long time, but every day brings new challenges and opportunities. I constantly encounter unique issues that require me to expand my knowledge and provide exceptional support to our valued customers.

    What surprises me is that no two tickets are ever the same, and hence, my job is never about sitting and answering the same questions repeatedly. Every customer has a different workflow, a different setup, and a different challenge they’re trying to overcome. Therefore, I get to put in conscious effort to investigate the issues and understand the unique workflow of each customer to suggest a suitable solution that meets their requirements.

    That’s what makes this support interesting and what makes me love doing it for all these years.

    At Fastmail, we’re committed to providing first-rate customer service around the clock through both our support form, and via email at [email protected]. As soon as we receive a ticket, our goal is to provide quick, helpful responses that resolve the issue as efficiently as possible.

    One thing that helps us tremendously is receiving clear and detailed information upfront. When customers include the relevant details in their first message, we’re often able to start troubleshooting immediately instead of spending time going back and forth gathering information. That can make a huge difference, especially when someone is already stressed about the issue they’re facing.

    Let us look at an example scenario to understand what kind of information can help in troubleshooting and resolving an issue faster.

    Emma is a long-time Fastmail customer who has been using the service for the past eight years to stay organised and manage her communication. One day, she noticed that she had not yet received a check from her insurance company. To follow up, she sent them an email asking for an update.

    After some time, she checked her Fastmail inbox expecting a reply. Instead, she found a bounce message. When she opened it, the error said: “Sender address rejected, SPF Failed, not authorised.”

    At first glance, it almost looked like the insurance company had not only failed to send the check, but had somehow blocked her emails as well!

    Emma knows that Fastmail Support is available to help, so she opens a support ticket with us. Since she’s already stressed about the issue, she’s not completely sure what details to include. She simply wants to let us know what’s happening and get help as quickly as possible.

    We know it’s not always easy to know what to share with us to make sure we can help you out quickly. That’s why we designed Fastmail’s support ticket form to be more user-friendly and straightforward.

    First, Emma entered her contact email address and name. If she had already been logged into her Fastmail account in the browser, she wouldn’t have needed to provide those details manually. We do encourage Fastmail customers to log in before submitting tickets because it helps us verify account ownership and safely discuss account-specific information when needed.

    Next, Emma selected the category that best matched her issue: “Mail, Contacts, Calendars.”

    She was then prompted to choose the type of issue she was experiencing, where she selected: “Trouble receiving or sending certain emails.”

    After that, she was asked for the subject, where she entered: “Email to my insurance company bounced.”

    A clear, descriptive subject line immediately helps us understand the issue and identify whether other customers may be reporting something similar.

    Because Emma selected a mail delivery issue, the form then asked for additional details such as:

    Sender
    Recipient
    Date, time and time zone
    Was a bounce message sent/received? If so, please copy and paste the error here.

    The form also included an area where Emma could describe the issue in her own words, and she responded appropriately to these straightforward prompts.

    Sometimes an issue doesn’t perfectly fit the options available in the form, and that’s completely okay. We encourage customers to describe the issue as clearly as possible and include any details they think may help.

    If an error message is involved, screenshots can also be extremely useful, especially if they include the full web page and URL. In many cases, understanding the exact steps that led to the issue helps us reproduce the problem more quickly. Screen recordings can be helpful too.

    Providing technical details about the device and environment can also make a difference, including:

    Device type
    Make and model
    Operating system version
    App or browser version

    After entering the information, Emma submitted the ticket.

    Shortly after submitting the ticket, one of our support agents will begin looking into the issue. In this case, I had the opportunity to assist Emma with her concern.

    As she had already included the sender address, the recipient address, and the complete bounce message as an attachment in her ticket, I was able to easily identify that she was having trouble sending an email to her insurance company and that the message had bounced back.

    The headers she included in the message indicated that she was sending the email using an external address through Fastmail’s mail servers. This resulted in an SPF mismatch, which is why the insurance company’s mail server rejected the message. In this situation, the bounce was actually expected behaviour.

    Since Emma had already shared all the important details upfront, I was able to quickly reassure her about what had happened and provide the correct steps to send emails from that address using the appropriate external mail server.

    If she had just said, “I can’t send emails,” I would have asked for those details right away. That would have delayed the resolution, which can be stressful, especially when you need a fast resolution as Emma did.

    Emma updated her settings based on my instructions and was successful in sending her message to her insurance company. She even sent us a lovely message to let us know that her concerns have been resolved. It was a delight assisting Emma!

    Emma’s ticket is a great example explaining what details are required for an easy resolution. A detailed support ticket with a clear subject line and relevant information is super helpful in fixing the issue. It also helps us spot trends in the ticket, allowing us to promptly involve the appropriate team and resolve the issue before it impacts more customers.

    One of the things we’re most proud of at Fastmail is the quality of personal support we provide to everyone — whether they’re long-time customers, trial users, or people simply considering the service.

    Over the years, we’ve received many thoughtful messages from customers, which gives us immense pleasure and truly adds to the satisfaction we find in helping others. I would like to share a selection of positive customer feedback from the extensive list of responses we have received. For the sake of brevity, only a few examples have been included below.

    “Thank you very much for the quick and excellent support. This is one of the reasons I love Fastmail. It took you only 12 minutes to reply again and specifically and perfectly answer every question I had on a rather complex topic. Stunning support, especially compared to what I experience with so many other companies. Definitely do share the praise with your team”.

    "I think I should really thank you for the quality of customer support you and your team do. Whenever I talk to fastmail, you always know what you are doing, what you are saying, and what I am saying. (I found out not every support@ does that, hard way, this week.) So,appreciate that.”

    "I just want to say, I’ve been impressed with the quality of the support responses I’ve received so far and how quickly the turnaround time has been. That’s always a good sign!”

    "Your friendly support is appreciated and is part of why we have kept Fastmail for all these years!"

    Choosing a few from the extensive list of customer feedback was not easy, and it was difficult to leave out many other thoughtful and encouraging responses.

    Every support ticket tells a story, and the more context you can share with us, the faster we can help bring it to a happy ending.

    If you’re ever not sure of what to share or what will help us troubleshoot, we’re always happy to walk you through that!

    If you’d like to troubleshoot the issue on your own, our help center is a great place to start, and our support form now offers optional AI Instant Answers. We have dedicated help pages for the most commonly reported issues, including sign-up, access, billing, mail client setup, DNS records, and much more. It is kept up-to-date and written by the same awesome support team that answers your questions every day. As soon as a feature is released, relevant help pages are uploaded for the customers. If you don’t find what you’re looking for, no worries! Just give us a shout, and we’ll be more than happy to lend a helping hand. We’re here to support you every step of the way!

    We hope this helps you get a speedy resolution the next time you need our support, and we’re looking forward to the next time we get to help you out!

    Original source
  • Apr 22, 2026
    • Date parsed from source:
      Apr 22, 2026
    • First seen by Releasebot:
      Apr 23, 2026
    Fastmail logo

    Fastmail

    An MCP server for Fastmail - National Email Day

    Fastmail releases an MCP server for AI clients, letting users connect Claude, ChatGPT and others to their mail, calendar and contacts through a new API with OAuth-based read, write and send access options.

    In brief

    For National Email Day, we’re shipping something new. Fastmail now has an MCP server at https://api.fastmail.com/mcp.

    MCP is an open standard, not tied to any one AI provider, for how AI clients talk to external data and tools. Think of it as another API sitting alongside IMAP, CalDAV, and CardDAV, except instead of being designed for mail apps or calendar clients, it’s designed for AI models to use directly.

    So if you point your AI client (like Claude or ChatGPT) at our MCP server, you can ask it things like “what’s on my calendar tomorrow?”, “draft a reply to Sarah’s email about the Q3 budget”, or “find my dentist’s address in my contacts” and it will do them against your real Fastmail account.

    The longer story - your data, your choice

    To be clear about what this isn’t: we have not integrated AI into Fastmail. There’s no chatbot bolted onto the inbox, and your mail isn’t being piped through a model in the background. The MCP server is simply another API endpoint for you to use, if you want to, with the AI client of your choice.

    That distinction matters to us. Our long-term values include “Your data belongs to you” and “We are good stewards of your data”. The pattern we try to follow is: rather than continuously reworking our UI to follow every new trend, we give you the interfaces to use your data however it suits you. MCP continues that pattern. It’s there if you want it, and nothing changes if you don’t.

    Using it is straightforward. Most AI clients (Claude, ChatGPT, and others) let you configure a list of MCP servers they’re allowed to access. You add https://api.fastmail.com/mcp, complete an OAuth consent screen to authorise the connection, and from then on your AI can read and act on your mail, calendar, and contacts when you ask it to.

    The OAuth consent screen will give you a choice of three levels of access: read-only (see emails, contacts, calendars), write (update emails, save drafts, edit contacts and events), and send (send emails).

    AI in your app, or your app in AI?

    When thinking about how webapps and AI integrate, there are two main ways of looking at it.

    The first (and the default today) is to integrate AI into the webapp. The webapp provider picks an AI system, builds it in, and ships it as a feature. A chat panel in the corner, a “summarize this” button, and so on. Each webapp ends up with its own AI, chosen by the developer, and none of them know anything about each other: your email has one AI, your project management system another, your document editor a third. From the user’s perspective, AI becomes a scattered collection of disconnected helpers, each confined to the webapp it shipped with and each missing the context that lives in the others.

    The second approach is to integrate the webapp into the AI itself. Here the user (not the app vendor) chooses the AI (maybe even local models run on just the user’s own devices), and that AI becomes a persistent interface that gets to know them over time. It can reach into whichever webapps the user has connected, pulling together context across email, projects, documents, and everything else and acting on their behalf. The webapp stops being a silo with its own bolted-on AI feature and starts being a system that the user’s AI can draw on.

    This second model is what MCP (Model Context Protocol) is designed to enable. It gives AI assistants a standard way to discover webapps, read data from them, and call functions on them, so that a single AI can orchestrate work across all the services a user has connected.

    The MCP Apps spec extends this further. It lets a webapp render its own UI inside the AI’s interface, so that interacting with a third-party service can feel native within the AI conversation rather than bouncing the user out to a separate tab.

    In that context, it’s not surprising then that MCP and MCP Apps are being driven by Anthropic, and other AI-only providers (like OpenAI) have also jumped on board as well. Meanwhile Google (at the time of writing) has no support in their Gemini chat UI for it, though there is support in some of the more developer-focused CLI tools. Google is simultaneously the ‘everything’ webapp, the AI provider, and the largest advertiser targeting you. They have an obvious incentive to favour the first model to keep you and your data within their own systems.

    Email - still the world’s largest communication platform

    Email is the world’s largest communication platform. It’s also open and federated across millions of providers rather than siloed inside a single organisation. That gives it a unique place in the world as your electronic memory that isn’t controlled by any one organisation. That also means everyone uses it in their own unique way, and Fastmail will continue to evolve to allow users to use their email, contacts, calendars and files how they want and where they want.

    Happy National Email Day!

    Original source
  • Mar 5, 2026
    • Date parsed from source:
      Mar 5, 2026
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Multi-window support and a better compose experience

    Fastmail releases a major interface update with multi-window composing, inline replies, a cleaner compose experience, reading pane placement below the inbox, signature controls, and desktop find-in-conversation.

    Today we’re releasing a major update to the Fastmail interface, with improvements focused on making mulitasking and compose better. Here’s what’s new.

    Multi-window support

    Sometimes you need to reference one email while replying to another, or draft several messages at once. Fastmail now supports opening compose and conversation views in separate windows.

    Hold Shift when clicking reply, forward, or any compose action to open it in a new window. You can also use Shift+R to reply in a new window, or Shift+A to reply all. Or pop out a whole conversation into its own window by shift-clicking in the message list, or using the button next to the subject in the top right if it’s already open.

    Your undo send timer works across windows too — if you send from a pop-out window, the undo notification appears in your main window so you’re still in control.

    Inline replies

    Replying to a message in a conversation now keeps you right where you are. Instead of switching to a separate compose screen, your reply appears inline within the thread, so you can see the full conversation while you write. It’s the most natural way to reply — you stay in context, and your draft lives alongside the messages you’re responding to.

    If you’d rather compose in a focused view like before, you can expand your reply to full screen at any time. We’ll remember your preference.

    A cleaner compose

    We’ve unified the look of our compose view across mobile and desktop, giving you a cleaner, more consistent interface wherever you’re writing.

    A few highlights:

    • Switch between Reply and Reply All mid-compose without losing your work. A new button in the compose header lets you change your reply mode on the fly, and Fastmail will intelligently recalculate the recipients for you.
    • Hide the formatting toolbar if you prefer a distraction-free writing area. A toggle in the toolbar lets you show or hide formatting options with a single click, and your preference is remembered.
    • Drag and drop recipients between To, Cc, and Bcc. The Cc and Bcc fields appear automatically when you start dragging an address, and tuck away again when you’re done if not needed. You can select multiple recipients using Shift-click for a range, or Cmd/Ctrl click to select individually.
    • Quickly add a recipient to your contacts or view their contact information if already a contact. Just click the recipient to select their token, then click again to get a menu of options.

    Move the reading pane below your inbox

    You’ve always been able to show a reading pane to the right of your message list. Now you can choose to place it below instead. This horizontal split gives you the full width of the screen for reading, which is especially useful on narrower monitors. You’ll find the option in your mail preferences.

    And more

    For those of us who don’t want our signature repeated in every thread, there’s now an option to turn off signatures on replies, forwards, or both. Head to Settings → Mail preferences to choose whether your signature appears above or below quoted text — or not at all.

    We’ve also added find-in-conversation to our desktop app. Just hit Cmd-F on Mac, or Ctrl-F on Windows/Linux, to get a search box letting you quickly find text in a long conversation.

    Original source
  • Oct 13, 2025
    • Date parsed from source:
      Oct 13, 2025
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Introducing the Fastmail desktop app

    Fastmail releases a dedicated desktop app for Mac, Windows, and Linux, bringing standalone convenience, offline support, native notifications, and default email client integration for a smoother, more focused mail experience.

    Fastmail is now available as a dedicated desktop app for Mac, Windows, and Linux. It’s the same Fastmail you know and love, now with the focus and convenience of a standalone app.

    With our desktop app you can:

    • Launch Fastmail from your dock or taskbar and find it in your platform’s app switcher.
    • Make Fastmail your default email client, so email links create a new message directly in Fastmail.
    • Work whenever, wherever, with full offline support, just like our mobile apps. You can always read your mail, manage your calendar, and write replies — your changes sync back seamlessly when you reconnect.
    • Whether you’re on macOS, Windows, or your favourite Linux distribution, you’ll find the app feels right at home on your platform, with native notifications, menus, and system integrations.

    Getting started is simple. Download the app for your platform, sign in with your Fastmail credentials, and you’re ready to go.

    Original source
  • Sep 17, 2025
    • Date parsed from source:
      Sep 17, 2025
    • First seen by Releasebot:
      Oct 15, 2026
    Fastmail logo

    Fastmail

    The Cache Crash

    Fastmail fixes a long-running test crash caused by file-locking issues after a fork, replacing the problematic process flow and adding safeguards in a new Cache::FastMmap open-source release. The change ends recurring test failures and helps protect other users from the same bug.

    Software Freedom Day and open source

    Software Freedom Day is this week, and we thought we’d mark the occasion with a post about some of our work on open source software that we maintain. Fastmail believes in being a good internet citizen, and that belief means we participate in the development of free software, both by sharing software that we write and by contributing to the maintenance of free software that we use.

    This is a highly technical post about how we solved a crash in our automated test systems. You do not need to make any changes to your Fastmail account or your email software, and you can ignore this post if it’s not interesting to you!

    The mysterious bug

    What happens when a mysterious bug causes all your tests to fail once or twice a week? We try to solve these mysteries, even if it takes a year.

    At Fastmail we have a strong testing culture around our code. When we write a new feature or fix a bug, we include tests to show that our feature works or the bug has actually been fixed. We open a Merge Request, and GitLab CI kicks off a job to run the project’s full test suite. We also have a scheduled job that runs once a day, testing our mainline branch. This checks that the tests haven’t started failing because of changes to the outside world.

    The job stands up a virtual machine, installs all of the necessary software, updates configurations, starts the various services that make up Fastmail, and then runs all of the tests. Finally, it shuts down the box and generates some downloadable artifacts with the test results. There are quite a few tests. All told, this process takes about 13 minutes. As a developer, you’re hoping to see this:

    File Count: 441
    Assertion Count: 22040
    --> Result: PASSED <--
    

    Starting some time in early 2023, we’d occasionally have jobs fail in strange ways. We didn’t see failures around the features we were changing. Instead, we’d see a bunch of random tests would fail with the error “Signal: 6” and “Signal: 11”:

    ( STDERR ) job 364 free(): invalid size
    < REASON > job 364 Test script returned error (Signal: 6)
    < REASON > job 364 No plan was declared, and no assertions were made.
    < REASON > job 369 Test script returned error (Signal: 11)
    < REASON > job 369 No plan was declared, and no assertions were made.
    < REASON > job 373 Test script returned error (Signal: 11)
    < REASON > job 373 No plan was declared, and no assertions were made.
    

    That means one process got a SIGABRT and then after that a bunch of processes got SIGSEGV. Something, somewhere, was going wrong with the software’s use of memory. “Use of memory”, though, covers a lot of what any given program is doing.

    At the time, our test runs didn’t have much logging, so we couldn’t really tell what was going on, and we couldn’t easily reproduce the problem. On any given run, either it happened or it didn’t. For a while, I just assumed our VMs were occasionally dodgy somehow. The problem was rare, so we tried to ignore it and just click “Re-run job” as needed.

    Soon though, it began to happen more frequently.

    I added logging to our builds, and I made sure that the job’s build artifacts would always contain all the logs. With this evidence in hand, we began to see a pattern:

    t1[43556]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    error 4 in FastMmap.so[7fca57fda000+6000] likely on CPU 2 (core 2, socket 0)
    t3[43580]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    t9[43685]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    error 4 in FastMmap.so[7fca57fda000+6000] likely on CPU 3 (core 3, socket 0)
    

    The crash was always in FastMmap.so! This comes from Cache::FastMmap, a Perl module of ours that provides an mmap-backed cache for use across processes.

    Armed with this information, I modified our builds to collect the on-disk cache files, too. With those, I’d be able to examine what was in the cache when a crash happened. Unfortunately, the cache files were corrupt, leaving me with more questions than answers.

    Tracking down the crash

    Over the course of months, I started taking a day here and there to dig through this problem – poring over the Cache::FastMmap code, our available logs, and our reproductions using the captured files. I tested and tossed out a lot of theories.

    Eventually, I added the ability to capture the core files generated during the test runs, because I realized that when this problem triggered, there was always a SIGABRT first with the error message seen below, which previously I hadn’t thought of as significant:

    ( STDERR ) job 364 free(): invalid size
    < REASON > job 364 Test script returned error (Signal: 6)
    

    It seemed pretty clear that the first crash was corrupting the file, and all of the processes using the cache file after were crashing because of that.

    So after adding the core file collector, I sat there clicking “Run job” on my merge request over and over and over and over and over, trying to trigger a crash.

    Eventually, it worked… but unfortunately I had no debugging symbols in the core files, so it didn’t help. I modified the build again, including those, and then clicked “Run job” a bunch more.

    Finally, I got what I needed, but this was even more confounding. The stack trace showed that the crash happened here:

    #3 0x00007f8c429d6195 in mmc_do_expunge (cache=cache@entry=0x56113b842070,
    num_expunge=<optimized out>,
    new_num_slots=<optimized out>,
    to_expunge=0x56113b506210) at mmap_cache.c:800
    800 free(to_expunge);
    

    to_expunge was a valid pointer, and nothing else had freed it yet from what I could tell by looking at the source.

    This told me that most likely we were writing over memory somewhere where we shouldn’t be, and we were stomping on malloc’s internal structures. Unfortunately, these kind of bugs are really hard to track down unless you can catch them in the act using tools like Valgrind or AddressSanitizer.

    I tried various things from there, including modifying the build to output every action taken against the cache file. The logs included a high-resolution timestamps, the action that was taken (fetch, store, delete), and whether it resulted in success or not.

    I replayed these recordings (single threaded - not a perfect test since it didn’t imitate inter-process locking or coordinating) with Valgrind and AddressSanitizer, but I got nowhere.

    I stared at the code, thought for a while longer, and then did some web searching for the “free(): invalid size” error to see if that could maybe help me pinpoint where a bad write was happening.

    That led me to a post on StackOverflow that pointed toward negative indexing. After reading that, I tried a few experimental programs until I reproduced the abort:

    #include <stdlib.h>
    void main(void) {
    int **x = (int**)calloc(5, sizeof(int*));
    int **y = (int**)calloc(5, sizeof(int*));
    int **z = x;
    int *num = malloc(sizeof(int));
    *--z = (int *)&num;
    free(y);
    free(x);
    }
    $ gcc negative-array-index.c
    $ ./a.out
    free(): invalid size
    Aborted (core dumped)
    

    Could it be that we were underflowing an array? That is, were we accidentally writing to memory below the lower bound of an array we’d allocated? With this theory in mind, I looked through the Cache::FastMmap source to find where we might be underflowing an array, and I spotted this loop in mmc_calc_expunge():

    MU32 ** copy_base_det = (MU32 **)calloc(used_slots, sizeof(MU32 *));
    MU32 ** copy_base_det_in = copy_base_det + used_slots
    //[... more code ...]
    /* Loop for each existing slot, and store in a list */
    for (; slot_ptr != slot_end; slot_ptr++) {
    MU32 data_offset = *slot_ptr;
    MU32 * base_det = S_Ptr(cache->p_base, data_offset);
    MU32 expire_on, kvlen;
    /* Ignore if if free slot */
    if (data_offset <= 1) {
    continue;
    }
    /* Definitely out if mode == 1 which means expunge all */
    if (mode == 1) {
    *copy_base_det_out++ = base_det;
    continue;
    }
    /* Definitely out if expired, and not dirty */
    expire_on = S_ExpireOn(base_det);
    if (expire_on && now >= expire_on) {
    *copy_base_det_out++ = base_det;
    continue;
    }
    /* Track used space */
    kvlen = S_SlotLen(base_det);
    ROUNDLEN(kvlen);
    ASSERT(kvlen <= page_data_size);
    used_data += kvlen;
    ASSERT(used_data <= page_data_size);
    /* Potentially in */
    *--copy_base_det_in = base_det;
    }
    

    This code allocates a new region of memory based on how many used slots we have (used_slots) and takes a pointer to the end of that region of memory:

    MU32 ** copy_base_det = (MU32 **)calloc(used_slots, sizeof(MU32 *));
    MU32 ** copy_base_det_in = copy_base_det + used_slots;
    

    Then, we loop over every slot in the cache, looking for ones that have data…

    for (; slot_ptr != slot_end; slot_ptr++) {
    MU32 data_offset = *slot_ptr;
    MU32 * base_det = S_Ptr(cache->p_base, data_offset);
    

    …skipping ones that don’t…

    /* Ignore if if free slot */
    if (data_offset <= 1) {
    continue;
    }
    

    …and finally copying over ones that do have data to the new region of memory, starting at the end of the memory region and working towards the front:

    *--copy_base_det_in = base_det;
    

    (Later, copy_base_det ends up in to_expunge seen above.)

    In order for this loop to underflow, the cache would have to have more slots with data in them than used_slots accounts for.

    The only way I could see this happening was if two processes wrote to the cache at the same time, corrupting its accounting. That, though, would mean locking had failed somehow…

    Finding the locking problem

    To chase that down, I modified the build to log when a lock was acquired, when a write happened, and when the lock was released. I clicked “Run job” wildly, like I was playing the world’s worst Cookie Clicker game… and eventually I got another hit.

    This time, there was a smoking gun:

    1739384152.734390 [32464] locked page 3: p_offset: 786432 pns: 179 pfs: 115 size: 262144
    1739384152.737327 [32928] locked page 3: p_offset: 786432 pns: 179 pfs: 115 size: 262144
    1739384152.738589 [32928] unlocked page 3 (with changes): pns: 179 pfs: 114
    1739384152.738776 [32464] unlocked page 3 (with changes): pns: 179 pfs: 115
    

    Here, pns is the number of slots, and pfs is the number of free slots.

    Above, we see that process 32464 locked the cache, and then before 32464 could write to the cache, process 32928 also locked the cache! That’s supposed to be impossible! Then, the second process wrote an entry to the cache, using up another slot, so it decremented the free slot count. And then, the first process wrote out its changes. Those didn’t include the use of any new slots, and so it set the free slot count back up to 115!

    This is what caused mmc_calc_expunge to underflow – it allocated memory for 64 slots (179 total slots minus 115 free slots) but ended up seeing 65 slots with data that it needed to copy over!

    Using the available logs on the machine, I identified the two processes, and found that one of them was a test that I had long suspected was the culprit.

    The cause and the fix

    That test uses IO::Async::Process to fork a child, then run some Perl code in that child. It turns out that IO::Async::Process closes all file descriptors except STDIN, STDOUT, and STDERR when forking. This is a surprising thing to do when forking – often a parent might intentionally leave file descriptors open for a child to continue using.

    A file descriptor is just a number representing an open file or socket (like a network connection).

    When Cache::FastMmap first starts up, it opens the cache file for reading/writing:

    int fh = open(cache->share_file, O_RDWR);
    if (fh == -1) {
    return _mmc_set_error(cache, errno, "Open of share file %s failed", cache->share_file);
    }
    cache->fh = fh;
    

    It then maps that file into memory:

    cache->mm_var = mmap(0, cache->c_size, PROT_READ | PROT_WRITE, MAP_SHARED, cache->fh, 0);
    

    Later, when it needs to write out changes to a specific page in the cache, it locks that region of mapped memory by locking specific ranges of bytes within the file using the file descriptor number:

    /* Setup fcntl locking structure */
    lock.l_type = F_WRLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = p_offset;
    lock.l_len = cache->c_page_size;
    /* Lock the page (block till done, signal, or timeout) */
    lock_res = fcntl(cache->fh, F_SETLKW, &lock);
    

    If another process using Cache::FastMmap with the same file tries to write changes, it will block trying to get the same lock until the current process is finished and unlocks the file.

    However, when IO::Async::Process closes all open file descriptors, cache->fh becomes invalid, since that number no longer belongs to any open file or socket. Then, later, if the child process opens up any new files or sockets, that number could be reused(!), so when Cache::FastMmap attempts to lock the cache in the child, it succeeds, but locks the wrong file!

    The memory mapped region is not tied to the file descriptor after having been mapped, so any changes written to memory would still overwrite the cache data, and stomp on whatever work any other process is doing at the same time.

    We fixed all this by ditching the “fork and run more Perl in the child” and replacing it with “fork and exec a brand new process”. With exec, we replace our running program entirely, so we no longer have the mmaped regions available. We’ve been segfault free ever since! We also added code to a new open source release of Cache::FastMmap, to try to prevent anybody else from going through all this.

    What we learned

    What’s really frustrating about this is that I almost cracked it months ago. I had suspected that the fork-and-close-children behaviour of IO::Async::Process was the problem, but when I tested some code manually to trigger the bug, I always got “invalid file descriptor” in the children during locking of the mmaped region… because my tests weren’t opening up new file descriptors. The lock file descriptor wasn’t valid and so the mmaped region couldn’t be be written over.

    Alas. be concise but fully comprehensive and precise.### Software Freedom Day and open source

    Software Freedom Day is this week, and we thought we’d mark the occasion with a post about some of our work on open source software that we maintain. Fastmail believes in being a good internet citizen, and that belief means we participate in the development of free software, both by sharing software that we write and by contributing to the maintenance of free software that we use.

    This is a highly technical post about how we solved a crash in our automated test systems. You do not need to make any changes to your Fastmail account or your email software, and you can ignore this post if it’s not interesting to you!

    The mysterious bug

    What happens when a mysterious bug causes all your tests to fail once or twice a week? We try to solve these mysteries, even if it takes a year.

    At Fastmail we have a strong testing culture around our code. When we write a new feature or fix a bug, we include tests to show that our feature works or the bug has actually been fixed. We open a Merge Request, and GitLab CI kicks off a job to run the project’s full test suite. We also have a scheduled job that runs once a day, testing our mainline branch. This checks that the tests haven’t started failing because of changes to the outside world.

    The job stands up a virtual machine, installs all of the necessary software, updates configurations, starts the various services that make up Fastmail, and then runs all of the tests. Finally, it shuts down the box and generates some downloadable artifacts with the test results. There are quite a few tests. All told, this process takes about 13 minutes. As a developer, you’re hoping to see this:

    File Count: 441
    Assertion Count: 22040
    --> Result: PASSED <--
    

    Starting some time in early 2023, we’d occasionally have jobs fail in strange ways. We didn’t see failures around the features we were changing. Instead, we’d see a bunch of random tests would fail with the error “Signal: 6” and “Signal: 11”:

    ( STDERR ) job 364 free(): invalid size
    < REASON > job 364 Test script returned error (Signal: 6)
    < REASON > job 364 No plan was declared, and no assertions were made.
    < REASON > job 369 Test script returned error (Signal: 11)
    < REASON > job 369 No plan was declared, and no assertions were made.
    < REASON > job 373 Test script returned error (Signal: 11)
    < REASON > job 373 No plan was declared, and no assertions were made.
    

    That means one process got a SIGABRT and then after that a bunch of processes got SIGSEGV. Something, somewhere, was going wrong with the software’s use of memory. “Use of memory”, though, covers a lot of what any given program is doing.

    At the time, our test runs didn’t have much logging, so we couldn’t really tell what was going on, and we couldn’t easily reproduce the problem. On any given run, either it happened or it didn’t. For a while, I just assumed our VMs were occasionally dodgy somehow. The problem was rare, so we tried to ignore it and just click “Re-run job” as needed.

    Soon though, it began to happen more frequently.

    I added logging to our builds, and I made sure that the job’s build artifacts would always contain all the logs. With this evidence in hand, we began to see a pattern:

    t1[43556]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    error 4 in FastMmap.so[7fca57fda000+6000] likely on CPU 2 (core 2, socket 0)
    t3[43580]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    t9[43685]: segfault at 8 ip 00007fca57fde1ee sp 00007ffdeab5ef00
    error 4 in FastMmap.so[7fca57fda000+6000] likely on CPU 3 (core 3, socket 0)
    

    The crash was always in FastMmap.so! This comes from Cache::FastMmap, a Perl module of ours that provides an mmap-backed cache for use across processes.

    Armed with this information, I modified our builds to collect the on-disk cache files, too. With those, I’d be able to examine what was in the cache when a crash happened. Unfortunately, the cache files were corrupt, leaving me with more questions than answers.

    Tracking down the crash

    Over the course of months, I started taking a day here and there to dig through this problem – poring over the Cache::FastMmap code, our available logs, and our reproductions using the captured files. I tested and tossed out a lot of theories.

    Eventually, I added the ability to capture the core files generated during the test runs, because I realized that when this problem triggered, there was always a SIGABRT first with the error message seen below, which previously I hadn’t thought of as significant:

    ( STDERR ) job 364 free(): invalid size
    < REASON > job 364 Test script returned error (Signal: 6)
    

    It seemed pretty clear that the first crash was corrupting the file, and all of the processes using the cache file after were crashing because of that.

    So after adding the core file collector, I sat there clicking “Run job” on my merge request over and over and over and over and over, trying to trigger a crash.

    Eventually, it worked… but unfortunately I had no debugging symbols in the core files, so it didn’t help. I modified the build again, including those, and then clicked “Run job” a bunch more.

    Finally, I got what I needed, but this was even more confounding. The stack trace showed that the crash happened here:

    #3 0x00007f8c429d6195 in mmc_do_expunge (cache=cache@entry=0x56113b842070,
    num_expunge=<optimized out>,
    new_num_slots=<optimized out>,
    to_expunge=0x56113b506210) at mmap_cache.c:800
    800 free(to_expunge);
    

    to_expunge was a valid pointer, and nothing else had freed it yet from what I could tell by looking at the source.

    This told me that most likely we were writing over memory somewhere where we shouldn’t be, and we were stomping on malloc’s internal structures. Unfortunately, these kind of bugs are really hard to track down unless you can catch them in the act using tools like Valgrind or AddressSanitizer.

    I tried various things from there, including modifying the build to output every action taken against the cache file. The logs included a high-resolution timestamps, the action that was taken (fetch, store, delete), and whether it resulted in success or not.

    I replayed these recordings (single threaded - not a perfect test since it didn’t imitate inter-process locking or coordinating) with Valgrind and AddressSanitizer, but I got nowhere.

    I stared at the code, thought for a while longer, and then did some web searching for the “free(): invalid size” error to see if that could maybe help me pinpoint where a bad write was happening.

    That led me to a post on StackOverflow that pointed toward negative indexing. After reading that, I tried a few experimental programs until I reproduced the abort:

    #include <stdlib.h>
    void main(void) {
    int **x = (int**)calloc(5, sizeof(int*));
    int **y = (int**)calloc(5, sizeof(int*));
    int **z = x;
    int *num = malloc(sizeof(int));
    *--z = (int *)&num;
    free(y);
    free(x);
    }
    $ gcc negative-array-index.c
    $ ./a.out
    free(): invalid size
    Aborted (core dumped)
    

    Could it be that we were underflowing an array? That is, were we accidentally writing to memory below the lower bound of an array we’d allocated? With this theory in mind, I looked through the Cache::FastMmap source to find where we might be underflowing an array, and I spotted this loop in mmc_calc_expunge():

    MU32 ** copy_base_det = (MU32 **)calloc(used_slots, sizeof(MU32 *));
    MU32 ** copy_base_det_in = copy_base_det + used_slots
    //[... more code ...]
    /* Loop for each existing slot, and store in a list */
    for (; slot_ptr != slot_end; slot_ptr++) {
    MU32 data_offset = *slot_ptr;
    MU32 * base_det = S_Ptr(cache->p_base, data_offset);
    MU32 expire_on, kvlen;
    /* Ignore if if free slot */
    if (data_offset <= 1) {
    continue;
    }
    /* Definitely out if mode == 1 which means expunge all */
    if (mode == 1) {
    *copy_base_det_out++ = base_det;
    continue;
    }
    /* Definitely out if expired, and not dirty */
    expire_on = S_ExpireOn(base_det);
    if (expire_on && now >= expire_on) {
    *copy_base_det_out++ = base_det;
    continue;
    }
    /* Track used space */
    kvlen = S_SlotLen(base_det);
    ROUNDLEN(kvlen);
    ASSERT(kvlen <= page_data_size);
    used_data += kvlen;
    ASSERT(used_data <= page_data_size);
    /* Potentially in */
    *--copy_base_det_in = base_det;
    }
    

    This code allocates a new region of memory based on how many used slots we have (used_slots) and takes a pointer to the end of that region of memory:

    MU32 ** copy_base_det = (MU32 **)calloc(used_slots, sizeof(MU32 *));
    MU32 ** copy_base_det_in = copy_base_det + used_slots;
    

    Then, we loop over every slot in the cache, looking for ones that have data…

    for (; slot_ptr != slot_end; slot_ptr++) {
    MU32 data_offset = *slot_ptr;
    MU32 * base_det = S_Ptr(cache->p_base, data_offset);
    

    …skipping ones that don’t…

    /* Ignore if if free slot */
    if (data_offset <= 1) {
    continue;
    }
    

    …and finally copying over ones that do have data to the new region of memory, starting at the end of the memory region and working towards the front:

    *--copy_base_det_in = base_det;
    

    (Later, copy_base_det ends up in to_expunge seen above.)

    In order for this loop to underflow, the cache would have to have more slots with data in them than used_slots accounts for.

    The only way I could see this happening was if two processes wrote to the cache at the same time, corrupting its accounting. That, though, would mean locking had failed somehow…

    Finding the locking problem

    To chase that down, I modified the build to log when a lock was acquired, when a write happened, and when the lock was released. I clicked “Run job” wildly, like I was playing the world’s worst Cookie Clicker game… and eventually I got another hit.

    This time, there was a smoking gun:

    1739384152.734390 [32464] locked page 3: p_offset: 786432 pns: 179 pfs: 115 size: 262144
    1739384152.737327 [32928] locked page 3: p_offset: 786432 pns: 179 pfs: 115 size: 262144
    1739384152.738589 [32928] unlocked page 3 (with changes): pns: 179 pfs: 114
    1739384152.738776 [32464] unlocked page 3 (with changes): pns: 179 pfs: 115
    

    Here, pns is the number of slots, and pfs is the number of free slots.

    Above, we see that process 32464 locked the cache, and then before 32464 could write to the cache, process 32928 also locked the cache! That’s supposed to be impossible! Then, the second process wrote an entry to the cache, using up another slot, so it decremented the free slot count. And then, the first process wrote out its changes. Those didn’t include the use of any new slots, and so it set the free slot count back up to 115!

    This is what caused mmc_calc_expunge to underflow – it allocated memory for 64 slots (179 total slots minus 115 free slots) but ended up seeing 65 slots with data that it needed to copy over!

    Using the available logs on the machine, I identified the two processes, and found that one of them was a test that I had long suspected was the culprit.

    The cause and the fix

    That test uses IO::Async::Process to fork a child, then run some Perl code in that child. It turns out that IO::Async::Process closes all file descriptors except STDIN, STDOUT, and STDERR when forking. This is a surprising thing to do when forking – often a parent might intentionally leave file descriptors open for a child to continue using.

    A file descriptor is just a number representing an open file or socket (like a network connection).

    When Cache::FastMmap first starts up, it opens the cache file for reading/writing:

    int fh = open(cache->share_file, O_RDWR);
    if (fh == -1) {
    return _mmc_set_error(cache, errno, "Open of share file %s failed", cache->share_file);
    }
    cache->fh = fh;
    

    It then maps that file into memory:

    cache->mm_var = mmap(0, cache->c_size, PROT_READ | PROT_WRITE, MAP_SHARED, cache->fh, 0);
    

    Later, when it needs to write out changes to a specific page in the cache, it locks that region of mapped memory by locking specific ranges of bytes within the file using the file descriptor number:

    /* Setup fcntl locking structure */
    lock.l_type = F_WRLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = p_offset;
    lock.l_len = cache->c_page_size;
    /* Lock the page (block till done, signal, or timeout) */
    lock_res = fcntl(cache->fh, F_SETLKW, &lock);
    

    If another process using Cache::FastMmap with the same file tries to write changes, it will block trying to get the same lock until the current process is finished and unlocks the file.

    However, when IO::Async::Process closes all open file descriptors, cache->fh becomes invalid, since that number no longer belongs to any open file or socket. Then, later, if the child process opens up any new files or sockets, that number could be reused(!), so when Cache::FastMmap attempts to lock the cache in the child, it succeeds, but locks the wrong file!

    The memory mapped region is not tied to the file descriptor after having been mapped, so any changes written to memory would still overwrite the cache data, and stomp on whatever work any other process is doing at the same time.

    We fixed all this by ditching the “fork and run more Perl in the child” and replacing it with “fork and exec a brand new process”. With exec, we replace our running program entirely, so we no longer have the mmaped regions available. We’ve been segfault free ever since! We also added code to a new open source release of Cache::FastMmap, to try to prevent anybody else from going through all this.

    What we learned

    What’s really frustrating about this is that I almost cracked it months ago. I had suspected that the fork-and-close-children behaviour of IO::Async::Process was the problem, but when I tested some code manually to trigger the bug, I always got “invalid file descriptor” in the children during locking of the mmaped region… because my tests weren’t opening up new file descriptors. The lock file descriptor wasn’t valid and so the mmaped region couldn’t be be written over.

    Alas.

    Original source
  • Aug 26, 2025
    • Date parsed from source:
      Aug 26, 2025
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Work offline with Fastmail

    Fastmail adds full offline support across its apps and webmail, letting users read, reply, manage contacts and calendar, and sync changes back when they reconnect. It also offers browser offline mode with settings controls and automatic rollout on mobile.

    The internet is available in more places every day, from subways to aeroplanes. But it’s still not universal, and it always seems to disappear at the most inopportune time. The overwhelmed mobile network cuts out just when you need that concert ticket. You realise you don’t have the address of your hotel just after landing in a country with no roaming agreement.

    Today, we’re pleased to announce full offline support for all our customers, in our apps and even on the web. No internet? No problem.

    How do I turn on offline support?

    We’ll be automatically enabling offline support for users of our iOS and Android apps progressively over the coming weeks. But if you can’t wait, or want offline support in your web browser, you can turn it on in Settings → Offline.

    Depending on the size of your account and the speed of your internet connection, it can take a few minutes or sometimes longer to do the initial sync. To allow the syncing to occur just leave the tab open in the browser (even in the background) until it shows it’s ready. Due to background processing restrictions on mobile platforms, our app can only do this initial sync while running in the foreground, but you can keep using it as normal while this is happening.

    By default, we’ll download the contents of recent messages, plus messages you open on your device. You can change this to all messages in the settings. Attachments are only cached for offline use when opened on the device.

    A few things to note if you want to enable offline support in your web browser:

    • You must tick “Keep me logged in” when logging in to be able to turn on offline support.
    • Offline support requires a modern browser — if you’re running something we can’t support, you’ll see a banner telling you this on the settings page.
    • Remember to bookmark your inbox to make it easy to get to. If you usually use a search engine to get to Fastmail, this won’t work without internet. A bookmark lets you open Fastmail’s webmail directly, which will load even when offline.

    What if I don’t want offline support?

    You can turn it off at any time in Settings → Offline. Turning it off will delete the local mail cache, so you will have to download it again if you change your mind.

    Is there anything I can’t do offline?

    We wanted to make the online-offline transition seamless, so you mostly shouldn’t need to think about it. Almost everything you can do online you can do offline, such as reading mail, replying, viewing and editing your contacts or calendar, and changing most settings. As soon as you’re online again, it will all sync back to the server.

    There are however a few minor differences to be aware of when working offline:

    • Mail search will not look inside attachments, and will give slightly different results to when online. And of course if you don’t choose to make every message available offline, it won’t be able to match against content it hasn’t downloaded!
    • Snoozed messages will not move back to the inbox while offline.
    • Calendar reminders will not show a notification while offline.
    • You can’t delete attachments from a message you’ve received.
    • You can’t add or change users or domains, change your plan or update your billing details, or change your security settings.

    What’s the tech behind your offline support?

    Interested in how we made this all work? We wrote up the technical details of the general architecture, how we sync changes back to the server, and how we made offline email fast when we launched the public beta late last year.

    Fastmail remains at the cutting edge of web development, with one of the fastest and most sophisticated apps anywhere on the internet. We’re super proud of this huge step forward in functionality, and our biggest hope is you almost don’t notice it — Fastmail now just works wherever you need it to.

    Original source
  • Aug 5, 2025
    • Date parsed from source:
      Aug 5, 2025
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Better themes, better navigation, better search

    Fastmail releases a major update to its web and mobile apps with refreshed theming, faster navigation, smarter search refinement, easier undo and redo access, and simple plain text editing in Fastmail Files.

    Today we released another significant update to the Fastmail web and mobile apps, introducing more beautiful theming and faster navigation on desktop, plus a powerful new search tool to help you find the email you’re looking for.

    Beautiful new themes

    We’ve revised the Fastmail UI to make it more balanced, more focused, and pull in more color from your theme. Head over to the settings and you’ll find many beautiful new default themes to choose from, or make it personal with custom colors.

    Faster navigation

    The tab bar in our mobile app has allowed switching between your mail, calendar, and other apps in just one tap for years. Now, webmail gets quick app switching too. Just like on mobile you can choose which apps to show in the settings, or turn them all off to hide the navigation bar entirely.

    For our customers with multiple accounts, it’s now much faster to move between them with our new quick user switching in the top right. Log in to more than one account in our mobile app and you’ll see the quick user switcher appear there too.

    Finally, our calendar on desktop now has faster navigation between day/week/month view. Want something more unusual? Click on the selected tab again to customise the number of days/weeks on show.

    Fine-tune your search

    Our new search refinement toolbar lets you quickly add, remove, or change filters to narrow down your search.

    We already have intelligent autocomplete to start you off, but sometimes the first attempt doesn’t quite give you the results you’re looking for. Our new toolbar makes it faster to quickly add useful filters to find the email you want.

    Have you ever done a search and found your results are being dominated by a particular sender you’re not interested in, or from a folder you know doesn’t have what you need? We intelligently offer to exclude these with just a few clicks, based on the previous search results.

    We hope you find this useful in getting the most out of Fastmail’s powerful search.

    And more

    Mistakes happen. That’s why Fastmail has multi-level undo/redo support for almost every action. But after the initial notification disappeared, you had to know the keyboard shortcut. Now you can undo/redo from the actions menu in your inbox as well.

    Fastmail files makes it easy to host a basic website. Now we’ve added simple editing of plain text files, so you can make basic changes without leaving the app, or create a new text file on the go.

    Original source
  • Jun 13, 2025
    • Date parsed from source:
      Jun 13, 2025
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Introducing the twom database format

    Fastmail introduces twom, a new Cyrus IMAP database backend with MVCC repacks, xxHash, mmap reads and writes, starvation-free locking, and faster pre-allocated file growth. Fastmail says all email servers have used it since February 12, 2025.

    I wrote back in December about my ideas for a new skiplist based database format for the Cyrus IMAP server.

    I spent a bunch of time over the next month writing just that. It’s called twom, “two” for the dual level0 pointers (the same as twoskip) and “m” for mmap and mvcc. I had planned to pronounce it “tomb” (for “it has tombstone records”) but have wound up usually saying “two em” for clarity. Maybe as this is posted on a Friday 13th we can call it “tomb” just for today.

    All the code was in a merged pull request. The database itself is just a single standalone C language source file and associated header file. There’s also a cyrusdb interface wrapper, and a copy of xxHash directly in our source tree as well:

    brong@elg:~/src/cyrus-imapd$ wc -l lib/*twom* lib/xxhash.h
    421 lib/cyrusdb_twom.c
    3298 lib/twom.c
    157 lib/twom.h
    7091 lib/xxhash.h
    10967 total
    

    Amusingly, xxhash.h takes more space than anything else (and the lion’s share of the CPU usage as well, despite being much faster than crc32 as used in twoskip).

    Why a new database format?

    As I wrote during our advent series, twoskip has served us well, but it had a couple of major performance issues - particularly with our current ZFS on NVMe architecture. Retro-fitting some improvements to twoskip would have been possible, but it couldn’t do the most important thing (MVCC reads) because the on-disk format didn’t have the right structure.

    We wanted MVCC because you could then repack an entire database without holding a lock the whole time, and replay the log at the end. I discovered that I didn’t even need an exclusive lock at all to do a repack, it could all be done with short-lived readlocks.

    Overall these changes made twom faster. We don’t have performance data to show how much faster because we changed too many things at the same time to isolate it, but one simple repack test of a giant file showed repacks that had been 35 minutes with the twoskip file taking just over a minute with twom. A rather massive improvement! And even better, you could write to the file during the repack without losing data, while twoskip would have held an exclusive lock the entire time.

    These are the major changes:

    xxHash

    The performance of xxHash vs CRC32 over small amounts of data is much better.

    Twoskip and twom formats both hash blocks of about 40 bytes for the tracking pointers, and our keys and values are quite short in a lot of Cyrus formats too.

    A faster hash function for small amounts of data is a big win. We chose xxHash for its friendly license and great performance.

    MVCC repacks

    This is massive. We have databases in the tens of gigabytes on the largest accounts, and when one of those chose to “checkpoint” - rewrite to remove stale data, it could lock an account for 30 minutes. This is obviously unacceptable.

    The same repack taking one minute and allowing reads plus a thousand or so opportunities for writes during that repack is a completely different story; speaking of which…

    Starvation-free locking

    I can’t believe I got all the way to testing this thing to discover how little I knew about fcntl locking.

    TIL: fcntl isn’t fair. Releasing the lock didn’t magically let a waiting writer proceed, the same process would often pick up the lock again without giving another process a chance to. On busy files, writers could entirely starve.

    We decided to use a two-offset locking strategy within the single database file, so writers can queue up waiting while all the readers are busy, and be ensured of their place in the line.

    The git history will show I didn’t get this right the first time, and had to do a patch while testing across our fleet on just the statuscache file, an ephemeral database with high churn. The great thing about testing with statuscache is that users won’t notice if it breaks, since any error will just cause the status to be re-computed!

    MMAP for reading and writing

    Twoskip (and all the other Cyrus internal formats, like cache and index) uses mmap for read but write for writes. On most operating systems this is fine, they share a common cache, but it seemed simpler and nicer to use mmap for both reads and writes rather than creating in-memory structures to copy over.

    And yes, we did read and watch the arguments against it (video)!

    Of course we’re using msync to get reliable commits, and twom still has robust transactions with the same 3-syncs-per-commit pattern that made twoskip so solid.

    MMAP also reduces syscalls. With twoskip we had to make multiple seek and write syscalls for each update, as we rewrote the backpointers (the average record has level 2, so needs to write 2 different backpointers plus the record itself).

    This means that a single twoskip transaction writing to a single key/value pair makes an average of 19 syscalls (2 fcntl, 2 fstat, 6 lseek, 4 write, 2 writev, and 3 fdatasync).

    Even with the more complex locking, the equivalent twom change makes half as many (4 fcntl, 2 fstat, and 3 msync).

    But even better - a transaction which updates multiple key/value pairs adds an additional average of 6 syscalls on twoskip (3 lseek, 2 write, 1 writev) per record, while twom has no additional syscalls per transaction, no matter how many changes are made. This is particularly obvious during a repack, where the initial transaction on the new file contains every record in the database!

    Pre-emptive allocation

    Every time twom needs to make the file bigger, it extends it by 25% and then fills in the empty space. This seems a good tradeoff between low numbers of truncate and mmap operations, while not making files insanely large. It doesn’t actually use that space on most filesystems, the file remains sparse until writes fill the space.

    This was by far the biggest performance increase (and the one we tested) - we had noticed with twoskip that a large amount of CPU was going on munmap and mmap calls, as with every new record the file became longer than the mapped space, and had to be re-mapped before the next write.

    Twom decouples the file size from the committed size, which can leave junk from aborted transactions on the tail of the file, but the header length never gets updated until the third msync, so nothing ever reads that junk, and it gets overwritten as new commits come in.

    Just straight POSIX

    The twom library is written to be standalone. It doesn’t use any of the supporting libraries from Cyrus, opting to do fcntl locking and mmap manipulation directly. This allows it to keep a list of active transactions with their own mmaps so pointers remain valid; and for many other optimisations. Many things from twoskip were tightened up and simplified by not relying on other libraries.

    This makes twom easily portable, and since it’s written from whole cloth (and I’m the only author on twoskip as well) I was able to put it under the CC0 public domain license; though obviously if you want to use xxHash you need to follow its 2-Clause BSD license as well.

    Twom databases are a single file, containing an ordered key-value list. It’s transactional, with single threaded exclusive writers and multiple parallel readers. Twom files can be accessed by multiple un-related programs concurrently, so long as they all obey the locking rules.

    I plan to lift the code out into its own repository at some point, and test it against all the other usual suspects in the key-value database space. The lib/cyrusdb_twom.c file is just a lightweight wrapper around the twom functions to convert cyrusdb semantics to twom style, and to convert error codes on the way back.

    OK what does it look like?

    Cyrus comes with a tool ‘cyr_dbtool’ which can be used to interact with any database formats, so here’s some interactions with a new DB, setting and deleting some records, and showing prefix iterators and dump output.

    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom set a b
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom show
    a b
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom dump
    UUID: uuid=fe720dec-4525-4e5d-a3c4-da665f3b0b40
    FNAME: fname=/tmp/test.db
    CHECKSUM ENGINE: XXH64
    HEADER: v=1 g=1 fl=10000000 num=(1/1) sz=(00000000/000001B0/00000170) ml=1
    00000060 DUMMY kl=0 dl=0 lvl=31 ()
    00000170 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000
    00000170 ADD kl=1 dl=1 lvl=1 (a)
    00000000 00000000
    00000198 COMMIT start=00000170
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom set xxa hello
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom set xxb world
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom show xx
    xxa hello
    xxb world
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom delete xxa
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom set xxa hi
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom dump
    UUID: uuid=fe720dec-4525-4e5d-a3c4-da665f3b0b40
    FNAME: fname=/tmp/test.db
    CHECKSUM ENGINE: XXH64
    HEADER: v=1 g=1 fl=10000000 num=(3/5) sz=(00000048/000002C8/00000170) ml=3
    00000060 DUMMY kl=0 dl=0 lvl=31 ()
    00000170 00000000
    000001F8 000001F8 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000
    00000170 ADD kl=1 dl=1 lvl=1 (a)
    00000280 00000250
    00000198 COMMIT start=00000170
    000001B0 ADD kl=3 dl=5 lvl=1 (xxa)
    000001F8 00000000
    000001E0 COMMIT start=000001B0
    000001F8 ADD kl=3 dl=5 lvl=3 (xxb)
    00000000 00000000
    00000000 00000000
    00000238 COMMIT start=000001F8
    00000250 DELETE ancestor=000001B0
    00000268 COMMIT start=00000250
    00000280 REPLACE kl=3 dl=2 lvl=1 (xxa)
    00000250 &lt;-
    00000000 000001F8
    000002B0 COMMIT start=00000280
    

    That’s a version 1 file, generation 1 (never been repacked), flags just means “using XXH64”, 3 commits, 5 records, some interesting sizes (last repack, current size, estimated repack size) - with the highest skiplevel of 3.

    Let’s repack it:

    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom repack
    brong@elg:~/src/cyrus-imapd$ /usr/cyrus/bin/cyr_dbtool -n /tmp/test.db twom dump
    UUID: uuid=fe720dec-4525-4e5d-a3c4-da665f3b0b40
    FNAME: fname=/tmp/test.db
    CHECKSUM ENGINE: XXH64
    HEADER: v=1 g=2 fl=10000000 num=(3/1) sz=(00000000/00000218/00000200) ml=2
    00000060 DUMMY kl=0 dl=0 lvl=31 ()
    00000170 00000000
    00000198 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
    00000170 ADD kl=1 dl=1 lvl=1 (a)
    00000198 00000000
    00000198 ADD kl=3 dl=2 lvl=2 (xxa)
    000001C8 00000000
    000001C8
    000001C8 ADD kl=3 dl=5 lvl=2 (xxb)
    00000000 00000000
    00000000
    00000200 COMMIT start=00000170
    

    All the tombstones removed, and the file is back in order. Any new writes will stitch themselves into the various linked lists by updating the back pointers. Finally, let’s look at the raw file:

    00000000 a1 02 8b 0d 74 77 6f 6d 66 69 6c 65 00 00 00 00 |....twomfile....|
    00000010 fe 72 0d ec 45 25 4e 5d a3 c4 da 66 5f 3b 0b 40 |.r..E%N]...f_;.@|
    00000020 01 00 00 00 00 00 00 10 02 00 00 00 00 00 00 00 |................|
    00000030 03 00 00 00 00 00 00 00 01 00 00 00 00 00 00 00 |................|
    00000040 00 00 00 00 00 00 00 00 00 02 00 00 00 00 00 00 |................|
    00000050 18 02 00 00 00 00 00 00 02 00 00 00 8a 57 92 ee |.............W..|
    00000060 01 1f 00 00 00 00 00 00 70 01 00 00 00 00 00 00 |........p.......|
    00000070 00 00 00 00 00 00 00 00 98 01 00 00 00 00 00 00 |................|
    00000080 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
    *
    00000160 00 00 00 00 00 00 00 00 6c d0 f8 56 00 00 00 00 |........l..V....|
    00000170 02 01 01 00 01 00 00 00 98 01 00 00 00 00 00 00 |................|
    00000180 00 00 00 00 00 00 00 00 d8 de f5 3d 84 a4 92 c8 |...........=....|
    00000190 61 00 62 00 00 00 00 00 02 02 03 00 02 00 00 00 |a.b.............|
    000001a0 c8 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
    000001b0 c8 01 00 00 00 00 00 00 05 8c 71 d7 70 cf ff c3 |..........q.p...|
    000001c0 78 78 61 00 68 69 00 00 02 02 03 00 05 00 00 00 |xxa.hi..........|
    000001d0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
    000001e0 00 00 00 00 00 00 00 00 19 cf b0 45 a7 69 06 9d |...........E.i..|
    000001f0 78 78 62 00 77 6f 72 6c 64 00 00 00 00 00 00 00 |xxb.world.......|
    00000200 07 00 00 00 00 00 00 00 70 01 00 00 00 00 00 00 |........p.......|
    00000210 1d 44 bd 3d 00 00 00 00 00 00 00 00 00 00 00 00 |.D.=............|
    00000220 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
    *
    00004000
    

    There’s not much fat in there! There is probably 50 bytes of overhead per record once you factor in trailing nulls on key and value, 8 bytes of header, 8 bytes of checksums, an average of 3 64-bit pointers, and padding out to an 8 byte boundary.

    Fastmail is running twom

    Since February 12, 2025, all Fastmail email servers have been using the twom backend in all the places they used to use twoskip. The switch was done in three phases over two days.

    I’m very happy to have removed one of the places in which Fastmail could fail to live up to its name! No more pauses for database repacks.

    I’m hoping that in some future release of Cyrus, the twom backend will be the default - but mostly, once twom was finished I was just glad to take a break from reading hexdumps and do something else for a while!

    Original source
  • Apr 28, 2025
    • Date parsed from source:
      Apr 28, 2025
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    A revamped Fastmail inbox

    Fastmail launches a new inbox look for webmail and mobile apps, with a cleaner message list, avatars, one-click hover actions, and easier customization. It also improves iPad input support, adds new mobile navigation and swipe options, and refines message actions.

    Today we’re launching a new look for your inbox, in both our webmail and mobile apps. Familiar but refined, the message list is now cleaner and more focused. Avatars help quickly identify who’s in the conversation, and one-click hover actions let you easily archive or delete.

    Make it yours with easier customisation, with live updates as you try out different options. Don’t have a Fastmail account yet? Here’s what you’re missing:

    And more

    We’ve also focused on fixing the niggling small things. We’ve improved support for using a mouse or pencil in our iPad app. Made it easier to quickly create a contact or event while reading a message. Added an option to stop messages from being automatically marked as read when you open them. We’ve even moved the “report phishing” action to be next to “report spam”. Finally.

    On mobile you can also now choose what apps are shown in the mobile navigation bar at the bottom (or turn them all off to hide it). Missing your quick hover actions? Configure up to four custom swipe actions to fly through your mail on the go. To select a message, just press and hold, or swipe with two fingers.

    We hope you enjoy the new look. Thanks to all our beta testers for the feedback they sent in. If you too want to get a sneak peek of what’s coming next at Fastmail, just log in at beta.fastmail.com.

    Coming soon … full offline support!

    Original source
  • Dec 16, 2024
    • Date parsed from source:
      Dec 16, 2024
    • First seen by Releasebot:
      Apr 4, 2026
    Fastmail logo

    Fastmail

    Dec 16: Offline support now in public beta

    Fastmail launches public beta for offline support in its app, letting users read mail, reply, manage contacts and calendar, and change settings without a connection. It also explains how to join the beta and what features still need internet.

    This is the sixteenth post in the Fastmail Advent 2024 series. The previous post was Dec 15: Platform Team working agreement. The next post is Dec 17: Building offline: general architecture.

    On an aeroplane, or down a tunnel. Roaming in a foreign land, or just out for a walk in the country. There are still times when we aren’t connected to the internet. As a strong supporter of open standards, you’ve always been able to use Fastmail with any email app you like, many of which work offline. However, many of our customers prefer the Fastmail app (we certainly do!), and offline support there has been our most popular request for some time. As Bron foreshadowed last year, we’ve been working hard on this big project for some time, and and we’re very pleased to announce it’s now ready for public beta testing.

    What’s a beta?

    Fastmail runs a beta version of our app to allow our users that like to live on the cutting edge to get early access to new features before they’re finished, and send us feedback while we’re still working on them. Things may be broken occasionally, but generally it’s pretty stable.

    How do I get the beta?

    If you are using our iOS or Android apps:

    Check for updates in the app store — make sure you have the latest version.

    Go to Settings → Device settings → Show Advanced Settings, and change the “server backend” to “Beta”.

    If you are using Fastmail on the web, just log in at betaapp.fastmail.com.

    In either case, once on beta you will find a new settings screen — Settings → Offline — where you can toggle on offline support. You must have ticked “Keep me logged in” when you logged in to enable offline support.

    What can I do offline?

    Almost everything! You can read mail, reply, view and edit your contacts/calendar, change settings, etc.…

    Probably easier is to describe what won’t work offline:

    Mail search will not look inside attachments, and will give slightly different results to when online. If you don’t choose to make every message available offline, it won’t be able to match against content it hasn’t downloaded!

    Snoozed messages will not move back to the inbox while offline.

    Calendar reminders will not show a notification.

    You can’t delete attachments.

    You can’t attach files to a message you are composing.

    You can’t add or change users or domains, change your plan or update your billing details, or change your security settings.

    How do I send feedback?

    Found a bug? Got an idea? Something you like, or don’t like? Let us know! Our support team will pass all feedback on to the product team. We promise we read and carefully consider it all.

    What’s the tech behind your offline support?

    Interested in the technical details? Over the next few days I’m going to dive into how we built offline support into our app, starting with the general architecture.

    Original source
Releasebot

Curated by the Releasebot team

Releasebot is an aggregator of official release notes from hundreds of software vendors and thousands of sources.

Our editorial process involves the manual review and audit of release notes procured with the help of automated systems.