[Super Search] Combining platforms Mac OS X and Linux gives no mac results
Categories
(Socorro :: Webapp, defect, P2)
Tracking
(Not tracked)
People
(Reporter: adrian, Assigned: willkg)
Details
Attachments
(3 files, 1 obsolete file)
| Reporter | ||
Updated•10 years ago
|
| Reporter | ||
Comment 1•8 years ago
|
||
Comment 2•7 years ago
|
||
| Reporter | ||
Comment 3•7 years ago
|
||
Comment 4•7 years ago
|
||
Comment 5•1 year ago
|
||
I was about to file a new bug and I found this one. The issue is still present, take this query for example. The "Platform" facet will only show Linux and Windows crashes. If we narrow it down to "Mac OS X" only tough we get a few crashes for that platform.
| Assignee | ||
Comment 6•1 year ago
|
||
Grabbing this to look at soon.
| Assignee | ||
Comment 7•1 year ago
|
||
There are a couple of problems.
First, I think the platform field is indexed wrong. It's indexed as an analyzed string, but should probably be indexed as a keyword.
Second, when there's only one value and it has a space in it, the super search code uses a simple_query_string query wrapped in a filter. If there's more than one value, then it uses a terms filter. These are completely different things. Specifying Mac OS X vs. [Mac OS X, Linux, Windows] creates two different searches and they behave differently. It's been this way since support for years and years. There's no explanation for why it was implemented this way or what the requirements were for the implementation.
We have to do some experimenting to figure out how to fix this in a backward compatible way with existing behavior for fields.
| Assignee | ||
Comment 8•1 year ago
|
||
| Assignee | ||
Comment 9•1 year ago
|
||
willkg merged PR [mozilla-services/socorro]: bug-1100790: fix super search platform queries (#6647) in 2df8d42.
On Friday, GitHub Actions weren't working, so I did other things.
Then today, I spent an hour writing a test harness that does this:
- fetches all the known supersearch fields
- for each field
- does a facet to get the top 3 values
- for each value
- does a supersearch using the default operator ('has terms" before this PR lands and "matches" after this PR lands) for 3 results and prints out the uuid and value
- does a superssearch using the default operator ("has terms" before this PR lands and "matches" after this PR lands) for all three values and prints out the uuid and value
This gives us a baseline of results before and after the change. It's hard coded to look at date >= 2024-06-24 < 2024-06-25 and at stage.
I'll run it before and after the stage auto deploy and compare the results. Depending on what happens, I'll either:
- cheer because everything is super duper!
- fix some additional things in followup PRs
- back this out because it needs some more thinking
| Assignee | ||
Comment 10•1 year ago
|
||
| Assignee | ||
Comment 11•1 year ago
•
|
||
I did another PR that fixes a bunch of other fields that are currently broken in prod and didn't get any better with my previous PR. After this second change, the following fields now all work with "match" operator:
crash_report_keysmac_memory_pressurexpcom_spin_event_loop_stackcrashing_thread_nameandroid_version
I wrote up bug #1905772 to fix android_model which is indexed poorly.
I didn't see any fields get worse with the "match" change, so I'm going to keep moving forward.
Before this goes to production, we should send out a heads-up about how the default operator is changing even if it's in a way that makes it more reliable and less flaky.
Comment 12•1 year ago
|
||
crashing_thread_name working correctly is going to be super-useful, thanks!
| Assignee | ||
Comment 13•1 year ago
|
||
This is my supersearch query test harness. I can't post the data because it includes protected data.
| Assignee | ||
Comment 14•1 year ago
|
||
I ran out of time. I'm backing out the first change and I'll redo that one and the second change when I get back from PTO.
| Assignee | ||
Updated•1 year ago
|
| Assignee | ||
Comment 15•1 year ago
|
||
| Assignee | ||
Comment 16•1 year ago
|
||
Comment 17•1 year ago
|
||
This was deployed in bug 1906174 just now.
| Assignee | ||
Comment 18•1 year ago
|
||
We reverted the changes for this bug, so no changes were deployed in 2024.07.03. I'll redo them this week.
| Assignee | ||
Comment 19•1 year ago
|
||
| Assignee | ||
Comment 20•1 year ago
|
||
willkg merged PR [mozilla-services/socorro]: bug-1100790: change supersearch terms query to matches (#6690) in 1c39163.
I need to notify stability and crashreporting-wg about this change before it goes to prod.
| Assignee | ||
Comment 21•1 year ago
|
||
I emailed stability and crash-reporting-wg mailing lists. I included examples of how the change affects queries. This fixes a ton of weird behavior.
| Assignee | ||
Comment 22•1 year ago
|
||
I went through and tested stage:
- supersearch seems to be working fine--I'm getting results I expect
- signature report looks fine--I'm not seeing any errors; results are what I would expect
- topcrashers report looks fine--I'm not seeing any errors; results are what I would expect
I think this is good to go to prod.
| Assignee | ||
Comment 23•1 year ago
|
||
This went to prod just now in bug #1915150. Marking as FIXED.
Description
•