Closed Bug 1466402 Opened 6 years ago Closed 6 years ago

Update livegrep to latest version

Categories

(Webtools :: Searchfox, enhancement)

enhancement
Not set
normal

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: kats, Assigned: kats)

References

Details

Lately indexing has been failing randomly while building the livegrep index, with the livegrep tool crashing. We should (a) update to the latest version of livegrep and (b) debug the crash. This bug is for (a).
https://github.com/staktrace/mozsearch/commits/newgrep The patch queue here updates our codesearch.py to use the grpc interface to talk to the codesearch daemon, and also makes us use the latest version of the livegrep repo. I did a reprovision with these changes (the AMIs end with -20180604 in the amazon console) and kicked off a dev run. It failed, as expected. But now I can move to step (b) which is debugging the problem. At first glance it looks like the index dumping step doesn't seem to write out the header information properly. So when the copy on the webserver tries to load the dump, it asserts at [1]. The dumping code [2] writes in a dummy header, then the body, and then goes back and fills the header in again with the correct values. So I'm guessing that last step isn't happening correctly. I'll try to reproduce this locally first since it'll be easier to debug than on the S3 instance. [1] https://github.com/mozsearch/livegrep/blob/cb5002239a50411f84ca86354723c01f7166a95c/src/dump_load.cc#L423 [2] https://github.com/mozsearch/livegrep/blob/cb5002239a50411f84ca86354723c01f7166a95c/src/dump_load.cc#L346
I wasn't able to reproduce the problem in my local vagrant instance :( I'll try debugging on the S3 indexer instance.
After much debugging, the problem turned out to be blindingly obvious and something I should have checked right at the beginning. The 100gb disk that we're writing to fills up! https://github.com/mozsearch/mozsearch/pull/113 to increase that disk size. I'll submit the PRs to migrate to the new livegrep anyway since I wrote those patches already, but will hold off for a day or two.
PR merged, although it hasn't been deployed due to bug 1483298
Status: NEW → RESOLVED
Closed: 6 years ago
Resolution: --- → FIXED
Assignee: nobody → kats
You need to log in before you can comment on or make changes to this bug.