I've been around for a really long time now.
Fri, Jun 23
Thu, Jun 22
Plus, I disagree with the assertion that we didn't do a full reindex. We did. Twice.
Why would the changes need to be reindexed if we're talking about accounts? This whole thing is stupid mess....
gerrit2@cobalt /var/lib/gerrit2/review_site$ java -jar bin/gerrit.war reindex --index accounts [2017-06-22 21:44:26,810] [main] INFO com.google.gerrit.server.cache.h2.H2CacheFactory : Enabling disk cache /var/lib/gerrit2/review_site/cache [2017-06-22 21:44:27,890] [main] INFO com.google.gerrit.server.project.ProjectCacheWarmer : Loading project cache Reindexing accounts: 100% (4569/4569) Reindexed 4569 documents in accounts index in 10.2s (445.8/s)
I don't need a date, reindexing accounts will take all of thirty seconds.
Ok, I guess we can reindex....again.
I mean not always, but it does most of the time I believe. In any case, we should still stop using OutputPage in maintenance scripts anyway.
Given that we've been unable to reliably reproduce this, I'm going to remove it from the 1.29 blockers for now.
Wed, Jun 21
I'm upgrading today to include the upstream patches for this, no need for deletion.
Thanks everyone for your help on this! Was quite a rabbit hole from the initial bug report :)
Hmm, all this started after we tried swapping SysV init for systemd. Funny how that correlates 🤔 😏
Fri, Jun 16
I'll have a look Monday
Updated on Gerrit as well.
Thu, Jun 15
That's not actually where it's timing out (yay buffered output and multi-threaded updates). It's actually wikidatawiki. But yeah, I'm on it.
Wed, Jun 14
NOT VENDOR DEFINITELY NOT VENDOR
Basically this bug needs me to undo the madness I started on in T137564.
This is totally related: T143188: "Key contains invalid characters" exception when purging pages with a "ä" in the title
Awesome progress, thanks!
I mean the performance gain over a get/set is pretty obvious. But if we can support it reliably we might need some sort of fallback.
As I said on IRC: why do we even attempt this codepath when we're not in binary mode?
Tue, Jun 13
Thanks for the quick fix!
I thought we shut this error up at least. It's back on wmf.5
Mon, Jun 12
Possibly. But we also deploy some extensions to beta that we don't branch for prod yet. I'd rather just stop using that meta repo entirely--can we think up another solution?
I'm not afraid of using it, I'm mostly concerned of how we'll handle beta. We currently have the extension meta repo symlink'd for easy installation. Sadly, this would make l10n regeneration impossibly slow if we only used --extensions-dir
Fri, Jun 9
"Note: In GitLab EE, you can configure multiple LDAP servers to connect to one GitLab server." - first thing that jumped out.
Sooo, I looked into Gitlab some time ago. It's a nice tool. The UI is certainly easy to use. Unfortunately, the feature list is incredibly lacking without buying the enterprise version--features that we'd absolutely need. For that, it basically became a non-starter in my book.
Looks better with the new UI
Thu, Jun 8
Per IRC, it's looking like these queries are coming from a malformed Bingbot request. So removed as blocker, although this is still pretty nasty and needs fixing.
Looks fixed, thanks!
Now it's fixed.
On re-reading the patch, it looks like $newRev does this, $oldRev still doesn't. I'll get a patch up.
The fix above ^ won't fix this at all. The error is that the function is being *passed* a null value, not that a null value is being *used*
Was backported and is now fixed.
I'm still seeing this in production.
Doesn't matter if it's one person or 300, cleaning up old patch sets is a pretty big pain in the ass :(
Kind of a pain if there's already patch sets :\
Nope, renaming is not possible :(
Wed, Jun 7
Rolled back because T152640.
Ugh. This came back with 2.13.8. @tstarling was unable to log in. After logging out, I could not log in again either. Casing didn't help, nor did reindexing accounts.
Tue, Jun 6
This should be fixed in all branches now. Resolve?
It would be an improvement if you could submit comments (including ones on older patch sets) and do +/- replies (which of course would apply to the top set) in one action instead of three actions.
Actually, this is fixed. You cannot begin editing from an older version of a change anymore.
This shouldn't actually be a problem anymore.