MediaWiki version: 1.35.0-wmf.38
PHP Fatal Error: Class MediaWiki\HookContainer\HookRunner contains 1 abstract method and must therefore be declared abstract or implement the remaining methods (MediaWiki\ChangeTags\Hook\ChangeTagsAllowedAddHook::onChangeTagsAllowedAdd)
Unknown, but this smells pretty broken.
A spike of 222 of these after deploy of wmf.38 to group1.
#0 [internal function]: MWExceptionHandler::handleFatalError()
#1 {main}| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | Release | brennen | T254175 1.35.0-wmf.38 deployment blockers | ||
| Duplicate | PRODUCTION ERROR | brennen | T256305 Fatal Error: Class MediaWiki\HookContainer\HookRunner contains 1 abstract method and must therefore be declared abstract |
This seems borked. HookRunner does have onChangeTagsAllowedAdd. Also, such a broken piece of code would have been spotted by several tests. Perhaps some internal PHP bug due to the many implements?
Unclear how this would have worked on group0. We will look into it further to see what we can learn.
All of these were on a single server (mw1287). Some sort of opcache issue or something?
Yes, very very likely T253673: Avoid php-opcache corruption in WMF production. Can you restart fpm and see if it goes away?
Mentioned in SAL (#wikimedia-operations) [2020-06-24T20:32:18Z] <cdanis> restarting php-fpm on mw1287 T256305
Mentioned in SAL (#wikimedia-operations) [2020-06-24T20:41:22Z] <brennen> train 1.35.0-wmf.38: attempting to roll forward to group1 after php-fpm restart on mw1287 (T256305, T254175)