It'd be great if fatals and exceptions were automagically logged into BZ with stack traces.
Obviously just logging every occurrence straight into BZ would be a mess.
There was talk of a logstash dedupe tool that might help allievate this problem...
Version: wmf-deployment
Severity: enhancement
See also: T85239
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | Feature | None | T70820 Automatically log fatals/exceptions into Phabricator with stack traces | ||
| Declined | None | T97136 Phabricator integration for Sentry | |||
| Resolved | • mmodell | T185155 Improve error reporting / integration between Kibana and Phabricator | |||
| Resolved | • mmodell | T230752 Deploy phatality into kibana |
Related: https://github.com/thurloat/sentry-jira
It lets you create JIRA issues from Sentry items/whatever they're called.
No Bugzilla plugin yet.
But there is a Phabricator plugin: https://github.com/getsentry/sentry-phabricator :) :)
So, that would allow manual task creation, not automatic. Maybe good enough for a first step.
A logstash to sentry output plugin (http://clarkdave.net/2014/01/tracking-errors-with-logstash-and-sentry/) may come in handy
Testing instance (un-puppeted POC) at https://sentry-beta.wmflabs.org. The phabricator plugin is installed. We need to 1) get some data in there, 2) setup a phabricator user to post bugz, 3) configure sentry to use the user.
For the "no private data" part, maybe we should setup the project that the bugs are posted to as private somehow and let a human triage them?
(In reply to Bryan Davis from comment #6)
For the "no private data" part, maybe we should setup the project that the
bugs are posted to as private somehow and let a human triage them?
That's what I was thinking, yeah. We can reuse what Mukunda did for the Security project, but no spam the Security project.
FWIW, Kibana X-Pack includes machine learning features which can be used to dedupe logs and identify outliers. If we implemented these features that might go a long way towards identifying which errors are unique and which ones should be automatically entered into Maniphest tasks.
More info: https://www.elastic.co/guide/en/kibana/5.5/xpack-ml.html
@mmodell it seems to me with Phatality deployed to production we can resolve this task ?
Phatality isn't fully automatic so I don't know if that addresses what @Reedy had in mind here or not.
An ideal tool would tell you whether a bug report for that stack trace exists already.
@Tgr: indeed, I want to add that functionality to phatality for sure. I've designed it so that will be possible, it just needs a bit more work to expose it into kibana's ui.