705796 - telemetry: gather stats on how often disk cache is corrupt.

Status: RESOLVED FIXED

(Core :: Networking: Cache, defect)

Description

[Jason Duell]

Attachment: v1 gather stats on how often disk cache is corrupt.

Like it says.  One nuance: don't gather a statistic (either false or true) if the cache is new (new profile/user): we want to be able to divide true/false to get the % of restarts where we see a corrupt cache.

Note that until this lands and we have data, SHUTDOWN_OK is probably a fairly good approximation.

Comment 1

[Bjarne (:bjarne)]

Comment on attachment 577314 [details] [diff] [review]
v1  gather stats on how often disk cache is corrupt.

Review of attachment 577314 [details] [diff] [review]:
-----------------------------------------------------------------

Weird..  didn't I review something like this some time ago? (Maybe I forgot to submit..?)

Anyway - I'd propose to combine this with the existing timer in OpenDiskCache() so that we get separate timing info for opening corrupted and working caches. The actual working/corrupted-ratio can be estimated by looking at the number of submissions (which will not be absolutely accurate, but with large enough numbers should give a nice indication).

If you don't want to combine them, the code is fine and r+. However, I'm clearing the review-request so that you can re-request if you decide to combine.

Comment 2

[Jason Duell]

Attachment: v2: unbitrotted

Michal,

I just found and unbitrotted this patch.  Do you think I should implement Bjarne's idea (if we did, I think we'd want to still keep this boolean metric so we know what % of the time the cache is corrupt), or just take this as it is?

Comment 3

[Michal Novotny [:michal]]

Comment on attachment 610758 [details] [diff] [review]
v2: unbitrotted

I don't think we really need to separate timings by working and corrupted caches. Time spent opening a corrupted cache should be always lower that in case of a working cache, so it won't affect the telemetry data in a bad way. I.e. high values in case of NETWORK_DISK_CACHE_OPEN should always represent opening a working cache. BTW ratio of opening corrupted and working cache should be pretty low and if it isn't we have a bigger problem than inaccurate NETWORK_DISK_CACHE_OPEN telemetry.

Comment 4

[Jason Duell]

https://hg.mozilla.org/integration/mozilla-inbound/rev/79662b6e372a

Comment 5

[Matt Brubeck (:mbrubeck)]

https://hg.mozilla.org/mozilla-central/rev/79662b6e372a