bingle-admin wrote:

The WMF core features team tracks this bug on Mingle card https://mingle.corp.wikimedia.org/projects/flow/cards/669, but people from the community are welcome to contribute here and in Gerrit.

It happened again. I tried to reply to topic http://ee-flow.wmflabs.org/w/index.php?title=Talk:Sandbox&workflow=050b9f659413f00fc8b1fa163e68c4ac
, it failed (probably due to bug 59195), and now if you try to view the topic you get this bug.

Werdna and I were able to reproduce this by reverting extensions/Flow on ee-flow back to

1bedba6... Merge "Repair missed conversion to new username lookup"

and replying to a post on User_talk:Werdna, which failed with bug 59195, in /var/log/apache2/error.log:

[Thu Jan 02 03:30:25 2014] [error] [client 76.14.58.232] PHP Fatal error:  Call to protected method Flow\\Model\\AbstractRevision::isAllowed()

When I then visited the post on User_talk:Werdna, it failed with this bug's error, in /tmp/debug.log:

[exception] [abe49e77] /wiki/User_talk:Werdna Exception from line 102 of /srv/mediawiki/extensions/Flow/includes/Data/RootPostLoader.php: Missing Posts: {"2":"050d44021127f00fc8b1fa163e68c4ac"}

I'm 90% sure the page showed this error even after switching git back to master.

Yet now, a few hours later without any DB changes, all these problem pages and topics on ee-flow are now viewable with no error. Hooray, but why?

beta labs has a similar page http://en.wikipedia.beta.wmflabs.org/wiki/Talk:Flow on which as I recall I made a failed reply, and that remains in an error state:

2014-01-02 05:27:21 deployment-apache32 enwiki: [d26ecf99] /wiki/Talk:Flow Exception from line 102 of /data/project/apache/common-local/php-master/extensions/Flow/includes/Data/RootPostLoader.php: Missing Posts: {"2":"050d408b42295cb204fa02163e0fb68f"}

Since the problem has vanished on ee-flow, I'm looking at labs now.
It's still throwing that same exception. Here's the full backtrace:

2014-01-02 15:58:38 deployment-apache32 enwiki: [41995562] /wiki/Talk:Flow Exception from line 102 of /data/project/apache/common-local/php-master/extensions/Flow/includes/Data/RootPostLoader.php: Missing Posts: {"2":"050d408b42295cb204fa02163e0fb68f"}
#0 /data/project/apache/common-local/php-master/extensions/Flow/includes/Block/TopicList.php(248): Flow\Data\RootPostLoader->getMulti(Array)
#1 /data/project/apache/common-local/php-master/extensions/Flow/includes/Block/TopicList.php(144): Flow\Block\TopicListBlock->getTopics(Object(Flow\Data\PagerPage))
#2 /data/project/apache/common-local/php-master/extensions/Flow/includes/View.php(72): Flow\Block\TopicListBlock->render(Object(Flow\Templating), Array)
#3 /data/project/apache/common-local/php-master/extensions/Flow/Hooks.php(151): Flow\View->show(Object(Flow\WorkflowLoader), 'view')
#4 [internal function]: FlowHooks::onPerformAction(Object(OutputPage), Object(Article), Object(Title), Object(User), Object(WebRequest), Object(MediaWiki))
#5 /data/project/apache/common-local/php-master/includes/Hooks.php(199): call_user_func_array('FlowHooks::onPe...', Array)
#6 /data/project/apache/common-local/php-master/includes/GlobalFunctions.php(4032): Hooks::run('MediaWikiPerfor...', Array)
#7 /data/project/apache/common-local/php-master/includes/Wiki.php(423): wfRunHooks('MediaWikiPerfor...', Array)
#8 /data/project/apache/common-local/php-master/includes/Wiki.php(305): MediaWiki->performAction(Object(Article), Object(Title))
#9 /data/project/apache/common-local/php-master/includes/Wiki.php(596): MediaWiki->performRequest()
#10 /data/project/apache/common-local/php-master/includes/Wiki.php(460): MediaWiki->main()
#11 /data/project/apache/common-local/php-master/index.php(49): MediaWiki->run()
#12 /data/project/apache/common-local/w/index.php(3): require('/data/project/a...')
#13 {main}

Timestamp for this ID 050d408b42295cb204fa02163e0fb68f, is 2014/01/01 23:28:15.

TopicListBlock::getTopics tries to load PostRevisions based on the return value of TopicListBlock::getPage: entries from TopicListEntry storage.

I've checked the database: flow_topic_list (TopicListEntry) and flow_revision (PostRevision) have no records of the id being fetched.

This would make it look like TopicListEntry cache is out of sync with DB (and purging the cache, letting it expire, or increasing its version number should "fix" that)

Workflowloader::commit, however, has some safeguard that should make sure that cache changes are buffered until DB transaction has been successfully completed, so I'm not yet seeing any reason for the cache to be having a value that is not in DB.

We think a patch fixed this.