Avoid decimating background process crash reports on release desktop
Categories
(Socorro :: Antenna, enhancement, P2)
Tracking
(Not tracked)
People
(Reporter: afranchuk, Assigned: bdanforth)
Details
Attachments
(1 file)
Like the gpu process (bug 1547804), other background processes (rdd, plugin, utility, socket) have very low submission rates (and crash rates, generally), and so they should likely always be accepted rather than being potentially decimated by the release firefox 10% rule.
|
||
Comment 1•1 year ago
|
||
There's an is_gpu rule already:
We could adjust that to be is_background and cover other background process types.
Here are counts by process type for the last week of Firefox 129.0.2 release channel:
$ supersearchfacet --_histogram.date=process_type --relative-range=1w --product=Firefox --release_channel=release --denote-weekends --format=markdown --version=129.0.2
| histogram_date | content | gpu | parent | plugin | rdd | socket | utility | total |
|---|---|---|---|---|---|---|---|---|
| 2024-08-28 | 7586 | 26 | 5367 | 9 | 14 | 0 | 47 | 13049 |
| 2024-08-29 | 7374 | 29 | 5332 | 16 | 0 | 0 | 36 | 12787 |
| 2024-08-30 | 6915 | 28 | 5422 | 13 | 3 | 0 | 28 | 12409 |
| 2024-08-31 ** | 5617 | 32 | 3812 | 12 | 0 | 0 | 29 | 9502 |
| 2024-09-01 ** | 5672 | 110 | 3763 | 14 | 6 | 0 | 0 | 9565 |
| 2024-09-02 | 7894 | 46 | 5495 | 9 | 0 | 0 | 0 | 13444 |
| 2024-09-03 | 7807 | 22 | 5413 | 25 | 6 | 1 | 81 | 13355 |
| 2024-09-04 | 4880 | 28 | 3417 | 11 | 0 | 0 | 0 | 8336 |
|
Assignee | |
Updated•1 year ago
|
|
|
Assignee | |
Comment 2•1 year ago
•
|
||
Similar to this comment for removing throttling of GPU process crashes, I wanted to provide an estimate of the expected increase in crash report volume if we no longer throttle crash reports for the (rdd, plugin, utility, socket) processes.
Since we throttle to accept only 10% of release Firefox crashes, we will assume that what we currently accept for these processes represents 10% of the total volume if we removed throttling.
For current release Firefox (130.0.0), there are 2715 crash reports from rdd, plugin and utility processes in the past week[1]. If we assume this week is representative of a typical week, then we can expect around 27k crashes per week from these processes once this is deployed in production.
[1]: Interestingly, I did not see any socket process crashes during this period, nor in the past 6 months. Is this expected?
|
|
Assignee | |
Comment 3•1 year ago
|
||
|
Reporter | |
Comment 4•1 year ago
|
||
(In reply to Bianca Danforth [:bdanforth] from comment #2)
[1]: Interestingly, I did not see any
socketprocess crashes during this period, nor in the past 6 months. Is this expected?
I do see some socket crashes in the past week, however I'm not filtering to 130.0.0, just the release channel. Did you change your filtering when looking at the past 6 months? Nonetheless, they are very infrequent. Those I see are single digits per day, if any. They are predominantly caused by third-party dlls injecting into the process and crashing :)
|
|
Assignee | |
Comment 5•1 year ago
•
|
||
Ah okay thanks! I was worried that there were none but I should have checked some different filters (or perhaps unchecked in this case) to open it up more. Glad you are seeing some.
|
|
Assignee | |
Comment 6•1 year ago
|
||
|
|
Assignee | |
Comment 7•1 year ago
|
||
We're hoping to get this change deployed to prod early next week. Once it's deployed, we'll add a comment to this ticket referencing the deploy bug.
|
|
Assignee | |
Comment 8•1 year ago
•
|
||
This was deployed in bug 1919125 tag v2024.09.16.
In the last 50 minutes since the deploy, there's been almost 100 crashes for these background processes compared to 4 in the 50 minutes before the deploy. That's on the order of the expected 10x increase, so it looks to be working!
Description
•