It is possible to make rollback fully automatic. In order to do so you need some integration with bootloader. It needs to be configured so that it can roll back to the previously working partition if update is broken. What is more, you can add some user space runtime checks that can verify the update and if those are not passing (updated image is broken) you can rollback to the previous one as well.
In my case, the first thing I do once an unverified version boots is to switch back to the other partition (so the known good version is active during the next boot), then run a detached reboot process that forces a reboot in 5 minutes. Once the system is up and it verified that everything is ok, it commits the next version (by switching back to the partition that booted and marking it as confirmed) so it is now active by default. Finally it kills the still running 'reboot' process.
As far as I understand your update process: You download a complete new version for every update and are able to stream that directly to the new partition? Is there any way to do delta updates? In my experience, most of the disk content is unchanged, unless you do major updates. In my case I download the new version using zsync, verify the downloaded/updated `install.zip` (which is kept on the volatile data partition), then extract that to the new partition. I make sure that `install.zip` is created in a way that it is rsyncable, so updates are pretty small that way. Of course you lose the streaming feature, unless you modify zsync somehow to support that.
The update is written to the inactive rootfs partition, uboot is configured to boot from it and the device is rebooted. Using the bootcount feature of uboot it is possible to roll back automatically if booting fails. Once the mender daemon comes up it will try to report the success of the deployment to the server. If this fails it will also roll back. Only after successfully reporting the success to the server Mender will "commit" the update, meaning configuring uboot to persistently boot from this updated partition.
Mender already does compression, but you are right that there are optimizations that can be made for application updates, e.g. delta or other types of updates. We are planning to implement this as well. The first priority for Mender is to make it robust, i.e. make sure the update is atomic and that you can always roll back.
I see. Thanks for the info. I suspected that u-boot does have support for that, but I wasn't sure.
> Once the mender daemon comes up it will try to report the success of the deployment to the server. If this fails it will also roll back.
Is there any deadline at all for that? I explicitly spawn a reboot command that ensures that even if everything gets stuck (in software, not in hardware) for whatever reason, the system falls back to the previous version (unless the reboot command gets killed too, in which case a manual restart is required). Any thoughts on that?
We have looked into hardware watchdog for this, but it is in the gray-zone of what an updater should be involved in. This is actually a more generic problem - maybe it hangs even when you did not deploy an update. There is varying support for hardware watchdogs across boards as well, unfortunately.
Most of the time it will not just hang, maybe it will crash or kernel panic and in those cases Mender will rollback. But the indefinite-hanging case is quite tricky and not yet handled.
Would be open to ideas here.
The first step of our boot process was to enable the watchdog. We extend the timeout periodically during the boot process, but generally if userspace isn't reached within 30 seconds or so we reset. Once in userspace, the daemon validates that things look good (this includes things beyond just application of the update -- did services start up correctly? Is the hardware operating as we expect?) before disabling the watchdog and marking the update as a success, at which point rollback isn't possible. At this point we might consider applying new updates, etc.
We also modified our first stage bootloader to be resilient to bootloader update issues, and chainloaded our second stage bootloader from a stub which could rollback.
We also niced the update process to avoid resource contention, allowed the updates to be delayed until the network was quiet, and paused them when it became noisy to make for a good user experience. There was a server-side flag to force updates to apply regardless, with higher priority, as well as one to basically disable all other functionality in the case of a unforeseen serious, perhaps security related, issue.
We actually had a discrete watchdog service which was responsible for petting an always-on watchdog, to rescue the system if it locked up or became unresponsive (if certain processes were not running, or responding, the watchdog would not be pet).
All of this led to effectively 0 failures in the field, a seamless user experience (except for the 30-second reboot when inactive). I wish everything I owned worked this way.
I could talk ad nauseum about this stuff. It's very cool to see the designs of others. I feel this is an under appreciated and under explored problem space.