[MDEV-9762] Make plugins/engines identify themselves in the error log upon startup and shutdown (on the Note level) Created: 2016-03-18 Updated: 2023-10-12 |
|
| Status: | Open |
| Project: | MariaDB Server |
| Component/s: | Plugins |
| Fix Version/s: | None |
| Type: | Task | Priority: | Major |
| Reporter: | Elena Stepanova | Assignee: | Unassigned |
| Resolution: | Unresolved | Votes: | 0 |
| Labels: | None | ||
| Description |
|
Currently there is no consistency in whether engines/plugins write a note about their start in the error log or not.
But others aren't so open about it. It becomes a major inconvenience when we investigate problems reported by users, because sometimes it's impossible to guess which plugins or engines were running at the time of failure; especially so if they were installed dynamically, so there is no sign of them in the cnf files (or if they were removed/commented out from cnf files since then). |
| Comments |
| Comment by Sergei Golubchik [ 2016-10-09 ] |
|
If you enable log-warnings with a level at least 3, you will see a note when a new plugin shared object is loaded |
| Comment by Sergei Golubchik [ 2016-10-09 ] |
|
May be it's better to list plugins in the crash report? |
| Comment by Elena Stepanova [ 2016-10-09 ] |
|
Level 3 is not very helpful in real-life situations – in bug reports I've never seen a user configuration having log-warnings=3, unless they are already in trouble and are investigating a problem; and in this case, there are other ways to get an up-to-date plugin list. If it's considerably easier to print the list upon a crash, and if it's reliable enough, it can be a way. By "reliable" I mean that upon a crash server is already in agony; if it's a choice/race between getting a plugin list and printing proper complete stack trace with all additional information it's already supposed to print, i would certainly prefer the latter. Printing it on plugin startup/shutdown just seemed more universal to me (and unified, as some plugins/engines already do it). |
| Comment by Sergei Golubchik [ 2016-10-10 ] |
|
I mean, how often do you have a complete error log in a crash bug report? Usually it's just a stack trace, and a few lines before and after, sometimes even too few, without a query that caused the crash. It's easier to log when a plugin is loaded and unloaded, than during the crash. But I wonder whether it'll help make crash bug reports any better… |
| Comment by Elena Stepanova [ 2016-10-10 ] |
|
We never have a complete error log on the first iteration, but we also hardly ever have config files and other things, so there are some iterations anyway. Requesting "server error log from the beginning of the last server startup" usually works. Please also note that it's not only (or always) about crashes, there can be other failures as well, e.g. server hangs, replication breaks, what not. I wouldn't of course mind having everything in a crash report as well, but I'm concerned about server's ability to do it. If it crashes even harder while trying to gather information about plugins, and cannot finish printing the stack trace and the guilty query, the loss will be much bigger than the gain. |
| Comment by Ralf Gebhardt [ 2023-10-12 ] |
|
I have removed the fixVersion for this task as it is one of 5.5, 10.1, 10.2, 10.3, which are EOLed. Please add a new fixVersion if you plan to work on this task for a not EOLed versions. |