[Bug] /usr/libexec/sailfish-osupdateservice osupdate-check sets ssu re

This bug report was filed in May 2019 at TJC, received a “Tracked by Jolla” by @jiit plus some additional explanatory edits by me in June 2019, and some details were discussed in its comments with @pvuorela in July 2019: [Bug] /usr/libexec/sailfish-osupdateservice osupdate-check sets ssu re

The very last statement from Jolla in July 2019 was:

Need to investigate if there’s any chance of things going wrong.

Well “things”, specifically /usr/libexec/sailfish-osupdateservice osupdate-check do go awry, as documented back then and observed many times before doing so (as mentioned there).

Because Together.Jolla.com has been put into read-only state more than a year after this little, initial conversation about this issue, and the promised transfer of content from TJC to this forum never materialised, this issue has to be re-filed here to be able to track and discuss it any further.

I still would appreciate any feedback on the content of this issue (in contrast to formal feedback and questions, which were already answered).

I also do not have the impression that this issue was ever addressed technically, at least I have found nothing mentioned in the release notes and changelogs of SailfishOS releases ever since.

P.S.: Cut & Paste of the core content of this issue.

[Bug] /usr/libexec/sailfish-osupdateservice osupdate-check sets ssu re

Tracked by Jolla

asked May 25 2019

olf gravatar image

olf
10393 ●104 ●201 ●158

updated May 6 2020

Table of content

Context

SailfishOS’ systemd user unit osupdate-check.timer regularly triggers the systemd user unit osupdate-check.service, which executes /usr/libexec/sailfish-osupdateservice osupdate-check.

Bug description

/usr/libexec/sailfish-osupdateservice osupdate-check sets SSU to the latest release or next “stop release” of SailfishOS, instead of solely checking, if a new release is available and then notifying the user.
The severe consequences do not immediately become apparent to the user:
All tools and applications relying on the package management might offer and incorrectly install RPMs for a newer SailfishOS release than the installed one, or they might fail to install RPMs due to incorrectly resolved dependencies.
Hence this affects the tools store-client, pkcon, rpm, libzypp and zypper supplied by Jolla, plus third party tools as Storeman / Warehouse (Openrepos client apps) and Patchmanager .
Letting any of these tools update all packages it manages then results in failing applications, SailfishOS not working properly or even a completely broken SailfishOS installation.

Resolution

Let /usr/libexec/sailfish-osupdateservice osupdate-check only perform the check and, if positive, emit an notification, but do not let it set SSU to a different release version than the installed one.
Consequently the output of version | grep -o '[0-9][0-9]*\.[0-9][0-9]*\.[0-9][0-9]*\.[0-9][0-9]*' | sed -n '1p' shall always be equal to the output of ssu re | rev | cut -f 1 -d ' ' | rev, unless the user explicitly commands otherwise.

Environment

Observed for many years on Jolla 1 phones and an Xperia X, each time I did not upgrade to the recent SailfishOS “GA (general availability)” release immediately. Finally I found the time to analyse this annoying issue.

To depict the trigger of the offender
[root]# systemctl --user status osupdate-check.service
● osupdate-check.service - Store client os update check
Loaded: loaded (/usr/lib/systemd/user/osupdate-check.service; static; vendor preset: enabled)
Active: inactive (dead) since Sat 2019-05-25 15:47:30 CEST; 1min 39s ago
Process: 7275 ExecStart=/usr/bin/invoker --type=silica-qt5 -n -d 5 /usr/libexec/sailfish-osupdateservice osupdate-check (code=exited, status=0/SUCCESS)
Main PID: 7275 (code=exited, status=0/SUCCESS)
[root]#

Workarounds

a. A quick fix, when the version displayed by ssu re differs from the output of version

Edit (2019-06-18): If you want to employ step b., do that before performing step a. (or redo a. after b.), because meanwhile the sailfish-osupdateservice osupdate-check may have been triggered again (until step b. is done)!

To resolve this issue until sailfish-osupdateservice osupdate-check is started again (10 minutes after a reboot plus every 7 hours 54 minutes), issue (as root)
ssu re "$(version | grep -o '[0-9][0-9]*\.[0-9][0-9]*\.[0-9][0-9]*\.[0-9][0-9]*' | sed -n '1p')"
(which sets SSU to the installed version), followed by a reboot.
Then either run post_sfos-upgrade (if sfos-upgrade is installed),
or manually execute the Final clean up steps” (specifically killing the store-client followed by removing its cached SailfishOS upgrade release version info),
plus a zypper ref (if zypper is installed)
and a pkcon refresh (i.e., always when doing this manually instead of using post_sfos-upgrade).

But if RPMs or Patchmanager Patches were updated or installed while SSU was set to a different SailfishOS release than the installed one, one should reinstall all these RPMs / Patches or (if these are untraceable) perform a factory reset of SailfishOS.

b. An additional, semi-permanent fix preventing sailfish-osupdateservice osupdate-check to be triggered by osupdate-check.timer

Note that performing a SailfishOS upgrade enables this timer unit again.
Thus redo step b. each time after upgrading Sailfish OS, until Jolla has fixed this (obviously not in SailfishOS 3.3.0 and apparently also not in any later version of SailfishOS as of August 2021).

A simple
rm /usr/lib/systemd/user/user-session.target.wants/osupdate-check.timer
(as root) followed by a reboot is fully sufficient for performing this step b.

4 Likes

I have ran into this same issue multiple times. After the OS detects the new upgrade being available, many “normal” operations become impossible, for example removing packages that should be removed before updating… IIRC, also installing new packages from OpenRepos is not possible due to dependency calculation going haywire.

I remember issuing “samegrade” upgrade and things got back to normal enough that I could then remove this and that software, and I then upgraded the OS using command line.

Simply a new OS upgrade being detected and available for install should not break the current system and its operations. That is just plain wrong.

2 Likes

Another way of dealing with this would be:

systemctl --user edit --full  osupdate-check.service

Add the following two lines to the [service] section:

EnvironmentFile=/etc/os-release
ExecStart=/usr/bin/invoker --type=silica-qt5 -n -d 5 /usr/libexec/sailfish-osupdateservice osupdate-check
ExecStartPost=/usr/bin/ssu re ${VERSION_ID}

This should automatically reset the version in a similar way olf proposes in the original post, while retaining the update check. You will find the new version of the service file in ~/.config/systemd/user.
Of course, systemctl --user edit only affects the execution under the user who does the edit. nemo/defaultuser in the majority of cases. Multiuser devices will require editing the system files after all.

2 Likes

Thank you for this enhanced workaround: More complicated to employ, but also more sophisticated.

Still I am expecting Jolla to fix this after having acknowledged this as a bug more than 3 years ago!
Your workaround may even be used as a “cover-up fix” (not addressing the real flaw, but masking its effects) by Jolla, if they would deploy it via a future SailfishOS release.

1 Like

Thank you for denoting that executing sfos-upgrade --verify might be used (after applying the initial workarounds to resolve mismatching release information, see above) to bring an already affected SailfishOS installation (i.e., which already has RPMs for a newer SailfishOS release installed, and you are unable to determine which ones) into a state, where one can fix things more easily, backup data etc. But as @direc85 denoted, the ultimate aim should be to upgrade to that newer SailfishOS release.

For a little detail on the --verify option of sfos-upgrade, see its description at Github.
sfos-upgrade is available at Openrepos, so it can be easily installed via Storeman.

P.S. / disclaimer: While I am the author of sfos-upgrade, the idea for the --verify option and its initial implementation was performed by @direc85. Hence all this is just shameless bragging from him and me! :wink:
More seriously, without him hinting here that sfos-upgrade --verify can be useful to rectify a SailfisOS installation to an extent, so one can perform system administration tasks and ultimately upgrade to “heal” the damage done by installing RPMs for a newer SailfishOS version (by upgrading to that version), I would have forgotten about this application of sfos-upgrade.

1 Like

An enhancement for “Workaround a.”:

  1. Issue ssu re "$(grep -F VERSION_ID /etc/os-release | cut -f 2 -d "=")" # Just another way to write the extant command above
  2. rm -f /var/cache/ssu/features.ini # New, guarantees together with step 3, that SSU uses the correct repos
  3. ssu ur # New, update repositories configured for SSU
  4. Then continue as described: Reboot, etc. etc.
1 Like

Thanks for moving this bug report from TJC to here @olf. We still have the internal report about this bug, so I’ve tagged it as “tracked” here too (and updated our internal report pointing to the new info you’ve provided here).

1 Like