Push events, NAT TCP connection timeouts, and device sleep

This is a technical post. Regular FastMail users subscribed to receive email updates from the FastMail blog can just ignore this post.

When we released the new user interface last year, one of the improvements included was push updates when new emails arrived.

In theory, push events are conceptually quite easy to do. We open a connection from your web browser to the server (see this blog post for details), then when a new email arrives, we send a message down the open connection to let the browser know. It then fetches the details about the new email(s) and refreshes the display.

Unfortunately, in the real world, it’s not quite that easy. The biggest problem is that when a mailbox is mostly idle (no new mail arriving), the connection from the browser to the server will be idle. While this shouldn’t be a problem, it turns out it often is.

As we have noted before, some of our users are behind NAT gateways/stateful firewalls that have short state timeouts. If you leave a TCP connection idle for too long (variable from 2 to 30 minutes depending on the device), these start dropping any new packets on the connection.

In the case of a push connection from the server to the client, this is particularly bad. When a new email arrives, the server will try and send data to the client, and then be told the connection is dead at that point. That’s fine for the server, it can then clean up the connection. However, the client will never see any data from the server, and neither will the client ever know that the gateway/firewall has broken the connection. The client will think it is still connected to the server and has no way of knowing that the connection has actually been broken. This is purely a consequence of the way the TCP protocol works. The only way for the client to be able to tell the connection is broken is to send some data down the connection, and there are only 2 ways that can happen.

  1. If the client has enabled TCP keepalive on the socket. Currently only Chrome on Windows does this.
  2. If the client sent some data down the connection to the server. Unfortunately the eventsource specification doesn’t provide any way to do this; it basically assumes the underlying TCP connection is always reliable and only the server can send to the client.

One way to try and work around this issue is for the server to send regular “ping” events to the client, sufficiently often that the gateway/firewall knows the connection is still alive. This is relatively straightforward to do, but causes other problems.

If the ping events come too fast, it can cause some clients to never go into sleep mode. For instance, we used to send ping events every 60 seconds. It was noted that on an iPad if you left the FastMail webpage open in Safari and put the iPad down, the iPad itself would never actually go to sleep. The screen would stay on, draining the battery very quickly.

Because of that, we decided to go the other direction and disabled the ping events, but that ends up back at the other end of the scale where sometimes push just seems to randomly stop working.

As there’s no perfect solution to this problem, we’re now changing again to a new trade off.

  1. The server will send regular “ping” events to the client at 5 minute intervals. This should be enough for most gateways/firewalls to keep the connection open, but long enough apart to allow devices to go to sleep.
  2. If the client doesn’t see a ping event after 6 minutes, it assumes the connection has died, drops the existing connection and creates a new one. This should at least allow push events to work to some extent on connections with gateways/firewalls with low timeouts.

This change has now been rolled out everywhere. Based on initial testing, we think that this time we’ve got the balance between theory and reality right.

Posted in Technical. Comments Off

Intermittent bayes db corruption resolved

This is a technical post that describes the history and recent efforts to track down a bug that was corrupting some users bayes databases. Fastmail users subscribed to receive email updates from the Fastmail blog can ignore this post if they are not interested.

Over the past few years, we’ve had sporadic reports of users bayes databases being corrupted and reset back to empty. When this happened, it would cause email delivery for that user to fall back to using the global bayes database, which decreased the overall accuracy of their spam detection until they retrained the database with more spam and non-spam messages.

I had tried multiple times to track down what was causing this issue, but each time with no luck. Each time the problem occurred, there was an error message in the logs of this form.

bayes: bayes db version 0 is not able to be used, aborting!

Often searching the internet for an error message will find other people that have had the same problem and tracked down the solution, but in this case it didn’t. Each time I tried to work through the code to see what was going wrong, I reached a dead end and couldn’t see any obvious problem.

Since the corruptions were very intermittent and losing a bayes database isn’t critical, doesn’t cause email to be lost or inaccessible, and can be rebuilt just by reporting email as spam/non-spam again, tracking this down was always a bit of a lower priority issue.

Recently though, after one more corruption report too many, I decided once and for all to track down what was causing it. Bit by bit over the course of several weeks, I added more and more logging information to the server code to track down where in the code the problem was occurring.

The logging results proved to be very odd. In the vast majority of cases it showed that writing to a particular database worked fine, but every now and then, it caused data to be lost. Eventually I managed to create a reproducible test case. It turned out to be very odd issue because performing a particular programming action with a database library worked fine the first 5 or 6 times, but on the 6th or 7th, it would cause data to become lost. Clearly something odd is happening in the lower level library code.

Fortunately there was a straight forward workaround to the problem, so I’ve now patched our code with the workaround, and over the last few weeks I’ve monitored the logs which show the original error message above has completely disappeared and no databases are being corrupted any more.

I’ve reported bugs to the underlying modules causing the problems, so hopefully long term they’ll fixed as well.

https://rt.cpan.org/Public/Bug/Display.html?id=83060

https://issues.apache.org/SpamAssassin/show_bug.cgi?id=6901

Posted in Technical. Comments Off

Update to DNS hosting

We’ve rolled out a change to our DNS hosting abilities to switch our backend from tinydns to powerdns. We’d previously tried this change once before but had some problems and had to roll back. After some more development work and testing, we believe we’ve fixed all these issues and so have moved forward to powerdns again.

This change should initially be invisible to users and things should continue to work as they were. In the long term, it will allow us to support more features and faster updates to DNS in the future.

Posted in Technical. Comments Off

Inter-tab communication using local storage

A few weeks ago we launched our new webmail service for all users at FastMail. Once being used by a wider audience, we of course received reports of a few edge cases our testing hadn’t managed to uncover. One of the more interesting issues we discovered came from this use case: our user liked to scroll down his inbox, opening each email he wanted to read in a new tab in the background. Then he would go through the tabs, closing each one as he was done with it. So far, so good. Except, in Chrome, his browser of choice, as soon as about 5 tabs were open, the rest failed to load, and the earlier ones then started having communication errors as well.

A quick bit of research and testing yielded the problem: Chrome limits itself to a maximum of 6 concurrent connections to a single origin across the whole browser. Each tab was loading a full instance of the mail application, which meant it was creating an EventSource object and connection to our push server, to be notified of new deliveries (see this previous post for how that works). Since these connections are permanent (that’s the whole idea!), opening lots of tabs quickly used up all the available connections, with none left to fetch any actual data. To the user, this appeared as “Could not connect to server” error messages.

The solution to this problem was not immediately obvious. Ideally, we would like to maintain a single push connection and share it between the tabs, but there’s no API for getting a reference to other tabs or windows in the browser, even if they’re pointed to the same domain. Then I remembered that setting a property on local storage triggers a “storage” event on the window object of every open tab with the same origin. This, I realised, could be used to synchronise behaviour across tabs.

The concept is fairly simple. Only one tab keeps a push connection; we call this the master tab. When it receives a push event, it broadcasts it by setting the event as a property on local storage called “broadcast”. When a tab receives the storage event for this key, it reads the JSON-encoded event object from local storage and processes it as though it had been received via an EventSource object.

The tricky part comes in coordinating between the tabs who should be master. The master tab also sets a value called “ping” on local storage roughly every 30 seconds to the current time stamp. When a tab first loads it checks for this value; if it is greater than 45 seconds ago it presumes there is no current master, so it becomes master. Otherwise, it becomes a slave. However, whilst it is a slave, it continuously monitors for storage events with a key of “ping”, and if it hasn’t heard a ping within a 45 second period, it takes over as master. This switches control to another tab when the master tab closes. On browsers supporting the “unload” event we can make the changeover happen pretty much instantly, by setting the “ping” value to 0 in local storage when the tab is closed.

This all works very well, but there’s one problem remaining: race conditions. There is no API for taking out an explicit lock on local storage, so the spec advocates the use of a per-origin mutex which would be acquired by scripts once they try to access the storage, and then released when the script finishes. Not all browsers have adopted this. The Chrome developers, for example, have decided the performance penalty is too great. Therefore, in some browsers, it is possible for scripts in different tabs to interleave such that, for example, each tries to take master at the same time, then each notices another has taken it so none end up as master! The solution we have adopted is to add a random component to the delay between pings and waiting for pings. This makes it unlikely that two tabs will both attempt to take master at the same time. Of course this can still happen, but should it do so, the random variation in each new master sending out a ping should ensure that one is quickly turned back to a slave. It will be eventually consistent, which is good enough for our purposes.

In case this is of use to anyone else, here’s the code we use (rewritten slightly to use pure JS rather than be based on our library code). It’s also available as a gist on github. You can try it out on this test page; just open the page in several windows or tabs, then close the master and see the control pass to another. You can also broadcast a message from any tab to the other tabs.

function WindowController () {
    var now = Date.now(),
        ping = 0;
    try {
        ping = +localStorage.getItem( 'ping' ) || 0;
    } catch ( error ) {}
    if ( now - ping > 45000 ) {
        this.becomeMaster();
    } else {
        this.loseMaster();
    }
    window.addEventListener( 'storage', this, false );
    window.addEventListener( 'unload', this, false );
}

WindowController.prototype.isMaster = false;
WindowController.prototype.destroy = function () {
    if ( this.isMaster ) {
        try {
            localStorage.setItem( 'ping', 0 );
        } catch ( error ) {}
    }
    window.removeEventListener( 'storage', this, false );
    window.removeEventListener( 'unload', this, false );
};

WindowController.prototype.handleEvent = function ( event ) {
    if ( event.type === 'unload' ) {
        this.destroy();
    } else {
        var type = event.key,
            ping = 0,
            data;
        if ( type === 'ping' ) {
            try {
                ping = +localStorage.getItem( 'ping' ) || 0;
            } catch ( error ) {}
            if ( ping ) {
                this.loseMaster();
            } else {
                // We add a random delay to try avoid the race condition in 
                // Chrome, which doesn't take out a mutex on local storage. It's
                // imperfect, but will eventually work out.
                clearTimeout( this._ping );
                this._ping = setTimeout(
                    this.becomeMaster.bind( this ),
                    ~~( Math.random() * 1000 )
                );
            }
        } else if ( type === 'broadcast' ) {
            try {
                data = JSON.parse(
                    localStorage.getItem( 'broadcast' )
                );
                this[ data.type ]( data.event );
            } catch ( error ) {}
        }
    }
};

WindowController.prototype.becomeMaster = function () {
    try {
        localStorage.setItem( 'ping', Date.now() );
    } catch ( error ) {}

    clearTimeout( this._ping );
    this._ping = setTimeout( this.becomeMaster.bind( this ),
        20000 + ~~( Math.random() * 10000 ) );

    var wasMaster = this.isMaster;
    this.isMaster = true;
    if ( !wasMaster ) {
        this.masterDidChange();
    }
};

WindowController.prototype.loseMaster = function () {
    clearTimeout( this._ping );
    this._ping = setTimeout( this.becomeMaster.bind( this ),
        35000 + ~~( Math.random() * 20000 ) );

    var wasMaster = this.isMaster;
    this.isMaster = false;
    if ( wasMaster ) {
        this.masterDidChange();
    }
};

WindowController.prototype.masterDidChange = function () {};

WindowController.prototype.broadcast = function ( type, event ) {
    try {
        localStorage.setItem( 'broadcast',
            JSON.stringify({
                type: type,
                event: event
            })
        );
    } catch ( error ) {}
};

Posted in Technical. Comments Off

The technology behind the classic and new interfaces

I recently wrote a postmortem for our old interface, now I want to explain how the addition of a modern interface alongside our classic interface is different.

In short, classic is here to stay.

For all that the interface has looked similar over the past few years, it’s had many changes under the hood.

Much of the interface is fully internationalised, both in classic and new. The code is all shared with the My Opera Mail product, where multiple language support is a core requirement.

It works much more nicely on small devices, in particular with Opera Mini.

Where possible, changes for the new interface have been rewritten as shared “library code” and integrated into both interfaces simultaneously. Some things (search, for example) work differently. But most core logic, and of course all low level mail routing and storage, are fully shared across our infrastructure.

This all adds up to a pile of “invisible” work we have done to make maintenance easier in future. Even the new search uses the same query builder library, so back-porting fully cross-folder search capability to classic will be achievable if there is demand.

Unlike the old interface, which was a completely separate copy of the code and grew stale over the years, there was never a “fork” (as it’s called in software development) for the new interface.

Indeed, you may have noticed that many screens on the new interface are really just “rebranded” classic. It’s the same HTML code as the classic desktop and mobile screens, with a different title bar. When you go back to the Mail or Address tabs, it reloads the javascript and hands control back. This was a deliberate decision to speed up the areas of our site where people spend 99% of their time (statistic taken from logs, not made up) without duplicating the rarely used screens. The client-side mailbox screen uses less bandwidth and is more responsive than the classic mechanism of downloading an entire html page on every click.

When we say “supported indefinitely” it really does mean that we have no plans to remove classic. There’s no internal timeline in our heads. The core technology is used by both interfaces, and we’re updating them together.

Finally to address concerns about continued IMAP access.

We have invested heavily on improvements to the Cyrus IMAP server, both myself and Greg Banks in the Australian office (who was hired to work full time on Cyrus, and is doing an awesome job).

Our new conversation features are built directly into Cyrus, and fully integrated into its replication system. Other features like storing previews and undelete information along with messages have been created by adding support to Cyrus for the standard message annotations described in RFC5257, contributing that work back to the community.

You can read more about the Cyrus project at http://cyrusimap.org/. This reliable and standards compliant server is the core of our technology stack. We’re not moving away from IMAP, even as we extend the server to support our specific use-cases.

You can read (or even download and play with) the exact code that runs on our servers from
http://github.com/brong/cyrus-imapd/ – our production systems run on the “fastmail” branch.

Bron.

Posted in Technical. Comments Off

Changes to delete behaviour with conversations

Over the past week we have changed how deletion works in our new modern interface. In this blog post I will explain what those changes were, and some actions we have taken to ensure no emails are accidentally lost.

This is a technical blog post, so it contains a moderate level of technical detail.

I will address how our backup and disaster recovery system works, and how we used it this week to recover emails which we suspected to be accidentally deleted.

Some background

Last week we rolled out the new conversations-enabled interface. However, we discovered we had under-estimated the impact of conversations on users’ existing workflow.

In particular, many users did not realise that when they selected a single item in a folder, it represented the entire conversation (all related messages, including those the user had sent).

When they pressed ‘Delete’ with one or more conversations selected, it deleted all messages in those conversations, including messages in other folders. For example, deleting a conversation in Inbox could also delete messages from “Sent Items” and “Important – Keep”.

We have altered the ‘Delete’ action to be safer in these ways:

  1. in a folder: only messages which in the same folder are deleted from the selected conversations.
  2. when viewing search results: only messages which match the search query are deleted, messages which are in the conversations but outside the search are not.
  3. when an action will cause more than one message to be deleted from a conversation, a warning message is shown to describe what will happen. The user must explicitly disable this warning if they don’t want to see it again.

What about the time before these changes?

Rather than leaving users to hunt for which emails were affected, We wrote a tool to data-mine our mail server logs. We log every create and delete of emails, along with enough data to identify which ones were “Delete to Trash”. We can also identify if the action came from an IMAP client or the web interface.

We found emails which could have been accidentally deleted using the following algorithm:

  1. the action came from the web interface.
  2. more than one message from the same conversation was deleted within 10 seconds.

All the emails which matched these criteria were restored back into the folders they were originally deleted from, with a custom keyword added. This makes it easy for users to find them again. Every affected user has been emailed with instructions on how to identify the restored emails.

How we restored data

When you delete an email on the FastMail servers, it isn’t immediately removed from disk, even if you manually expunge via IMAP. We do this:

  1. to guarantee that our “Restore from backup” feature can always find all your emails, even if they were delivered and then deleted in between backup runs.
  2. to make deletes appear faster to users.
  3. to reduce the load on our IMAP servers. Removing files is actually one of the slowest operations you can run on a modern filesystem.

So we actually batch up all deletes and run them once per week at the least busy time for our servers – Saturday night in the USA. It’s weekend everywhere in the world then.

We also never remove email files within one week after deletion, so that our “Restore” feature can work as advertised.

This is, of course, in addition to the safety provided by replication to an offsite datacentre, and daily backups to a different server running a different operating system.

Immediate response

As soon as we realised we may have to restore emails, we disabled the automated weekend cleanup job, and started collecting data from our servers.

Discoverability

The problem is that it is hard to know that an email is not there unless you actively look for it. We could disable cleanup temporarily, but not forever. Our turnover is about 2% of total email volume per week, so the disks would fill up if we never deleted anything ever again.

We decided the safe way forwards was to undo every deletion which had even the slightest chance of being by accident.

That way, if no action is taken, a few extra emails sit on disk gathering dust. It’s possible at any later time to discover them and clean them up. There is no requirement to act quickly.

We take your privacy very seriously. No contents of emails were accessed during this task, and each user’s account was processed separately to ensure there was no risk of disclosing data. You can read more about our privacy policy here: https://www.fastmail.fm/help/overview_privacy.html.

Data collection

We log every single time a message is added to or expunged from any folder on out backend servers. We collected an initial dataset of nearly 30 million “Delete to Trash” events from the log files.

The next step was identifying which of these were a single action involving more than one message from the same conversation. Every message was tagged with a session identifier and timestamp as well as the folder and IMAP “UID” which uniquely identifies it, but we were not logging the CID (conversation identifier). We do now, but that doesn’t help with log lines from the past!

Finding the CID involved writing custom code to read the index file on disk (which still contains the deleted record) and extract the CID field for every deleted message.

Finally of course, there was processing the logs for every single connection from the web servers over that time frame and finding which deletes were related to each other. There’s nothing in the log to show that it’s the same command, so we applied a heuristic of “within 10 seconds” to account for the outside case of a busy server and large folders being processed.

Restoring messages

We use the Cyrus IMAP server. One of the utilities included is called ‘unexpunge’, and it can be used to recover deleted emails. This is different from our usual restore command, which extracts messages from various sources and appends them a new temporary folder.

In this case we want to restore messages permanently, so unexpunge is the right tool… except – we want to tag every message with a keyword, and we want it to be reliable. Finding the messages afterwards is messy. We chose to add a new feature to unexpunge, setting a user-defined keyword on each message as it is restored. It is robust, and there’s no gap where messages appear without the keyword

The chosen keyword is RESTORED-20121107. Our web interface already supports global keyword search with “flag:$name”, so the email to users includes a pre-generated URL which will perform a global search on all that user’s folders for messages which were restored.

Restores are in progress now. Once they are completed, thousands of users will have some messages restored. This is almost certain to include messages which were intended to be deleted, but we must err on the side of safety here.

We have built a very robust infrastructure because of our strong commitment to data safety. These restores are in line with this commitment. It is easier to delete unwanted messages again than to recover messages which no longer exist.

Posted in Technical. Comments Off

New login and session management code on beta.fastmail.fm

We’ve just rolled out some new code on our beta server that significantly changes how sessions are managed. This new code reduces some overall session complexity, fixes some long term bugs, and adds some useful new features.

  1. There’s now just 2 main types of sessions: normal & long term
    • normal – these expire after 2 hours of inactivity
    • long term (you check the “Keep me logged in” checkbox on login) – these expire after 30 days of inactivity, for most people on most machines, this is effectively forever

    (Note: The "Keep me logged in" checkbox has been broken for the last few months on the beta server, but now correctly creates a long term login session. Also the "lightbox" login screen within the new UI now correctly works.)

  2. Logout will explicitly end a session 

    If you want to explicitly end a session, use the "Log out" link at the top right of the page. If you want to keep a session, just close the browser tab/window and when you go back to the beta server, you’ll still be logged in (see below).

  3. You can still log in to multiple different accounts

    We still support the ability to log in to multiple different user accounts at the same time on the same device/browser.

  4. You can access existing logged in sessions from the login screen

    If your device/browser has any existing logged in sessions, we now show those sessions when you go to the login screen. Simply clicking on one of those sessions will send you straight back to that mailbox for that user.

    Although by default the login screen shows existing logged in sessions, clicking the "Log in to another account" link will allow you to log in to another account at the same time.

  5. You can see (and remotely log out) all logged in web sessions on all devices/browsers

    We now track all sessions in our database and allow users to see all these sessions and remotely log out any of them individually.

    Just go to Options/Accounts –> Logged In Sessions to see all sessions in all devices/browsers. Currently only sessions created on http://beta.fastmail.fm can be deleted.

    (Note: Only web sessions are shown. IMAP/POP/XMPP/etc logins are shown on the Options -> Login Log screen)

One observation that some people might make is that with the old system, if you were logged into your account, and then closed your browser window/tab or went to http://beta.fastmail.fm again, it would appear that your existing session was automatically logged out, a nice security feature.

In fact that was never the case, the session was not logged out. Simply picking the right URL from your browser history would take you straight back in. There was just no visual indication on the login screen that this existing session was still present in your browser cookies, which is actually quite dangerous. The new system correctly shows any existing sessions on the login screen. If you want to end a session, you must use the "Log out" link at the top right of the page, whether you’re using the new system or the current system still at http://www.fastmail.fm.

Posted in News, Technical. Comments Off
Follow

Get every new post delivered to your Inbox.

Join 5,258 other followers