Raise Global Dispatcher Limit to 1_000_000
Categories
(Data Platform and Tools :: Glean: SDK, enhancement, P2)
Tracking
(Not tracked)
People
(Reporter: perry.mcmanis, Assigned: chutten|PTO)
References
Details
Attachments
(1 file)
Once we use an unbounded queue, it should be possible to raise the global dispatch rate to 20k as it doesn't have to be pre-allocated.
Once confirmed that use of an unbounded queue is acceptable and does not cause any regressions, we will test raising the limit, which should hopefully help to address pre-init overflow on FOG.
Updated•3 years ago
|
Assignee | ||
Comment 1•3 years ago
|
||
*address current pre-init overflow on FOG.
There are some changes waiting in the wings (notably bug 1790872) that'll require the limit to be even higher (or for a different solution to be found). I'd like to push for the limit to be configurable by embedders.
Yes, I realize that the main way embedders configure Glean is via init which is something that happens notably after pre-init. I'm not sure how best to configure this... cargo features would be a clumsy fit to try and tune this, maybe cargo config [env]
blocks and the option_env!
macro would work (or moz.build
, honestly) (but how do we ensure the env is set when building the Rust crates), we could add another API that can be called during I dunno static init or any other time before the queue overfills....
I dunno how it'd work. But without it I see me having to shepherd Glean SDK releases regularly as new pre-init instrumentation is added to Firefox Desktop.
Maybe FOG just goes full unbounded?
![]() |
Reporter | |
Comment 2•3 years ago
|
||
From SDK meeting: use an optional flag to conditionally set the global dispatch queue limit for the pre-init queue to 1 million on desktop
Assignee | ||
Comment 3•3 years ago
|
||
Assignee | ||
Comment 4•3 years ago
|
||
Assignee | ||
Updated•3 years ago
|
Description
•