Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: 0b589ff5c2f2066348731a0f48c271c63dd43ad7
https://github.com/WebKit/WebKit/commit/0b589ff5c2f2066348731a0f48c271c63dd43ad7
Author: Ben Nham <n...@apple.com>
Date: 2025-03-23 (Sun, 23 Mar 2025)
Changed paths:
M Source/WebKit/NetworkProcess/cache/NetworkCacheData.cpp
M Source/WebKit/NetworkProcess/cache/NetworkCacheData.h
M Source/WebKit/NetworkProcess/cache/NetworkCacheDataCocoa.mm
M Source/WebKit/NetworkProcess/cache/NetworkCacheStorage.cpp
M Source/WebKit/NetworkProcess/cache/NetworkCacheStorage.h
Log Message:
-----------
Stop using IOChannel in NetworkCache
https://bugs.webkit.org/show_bug.cgi?id=290211
rdar://142645876
Reviewed by Per Arne Vollan.
This is a speculative fix for a hard-to-repro issue where page loads sometimes
hang for minutes
while the system is under heavy load.
What we see in logs is that the disk network cache lookup ends up on a pri 4
default overcommit
dispatch queue. That thread is parked in a pread and gets preempted for minutes
while other higher
pri work is occurring. The thing that is enqueuing the pread on to that
dispatch queue is
`dispatch_io_read`, which enqueues that block on to a queue that it controls.
We think that probably
something is going wrong with the priority propagation to that queue sometimes,
as the network cache
lookup hops from main thread => network cache I/O queue => dispatch_io private
queue and back.
As a speculative fix:
- Stop using `dispatch_io` / `IOChannel` altogether. It buys us nothing other
than adding extra
thread hops, making things harder to debug, and perhaps exposing some latent
bug in priority
propagation. Instead, just do the I/O directly on the network cache's I/O
queues.
- Change the priority of the read-only I/O queue from default to
UserInteractive, since it's on the
blocking path for resource fetching.
- Change the priority of the write I/O queues from Background to Utility, since
Background is very
punitive (capped at max pri 4), and if we get anything wrong, getting capped
at max pri 4 can
cause minutes of starvation.
WriteOperations also had an completion handler that received an integer errno
as an argument. This
was unused, so I removed it.
* Source/WebKit/NetworkProcess/cache/NetworkCacheData.cpp:
(WebKit::NetworkCache::Data::Data):
* Source/WebKit/NetworkProcess/cache/NetworkCacheData.h:
* Source/WebKit/NetworkProcess/cache/NetworkCacheDataCocoa.mm:
(WebKit::NetworkCache::Data::Data):
* Source/WebKit/NetworkProcess/cache/NetworkCacheStorage.cpp:
(WebKit::NetworkCache::Storage::WriteOperation::WriteOperation):
(WebKit::NetworkCache::Storage::WriteOperation::~WriteOperation):
(WebKit::NetworkCache::Storage::Storage):
(WebKit::NetworkCache::Storage::dispatchReadOperation):
(WebKit::NetworkCache::Storage::readRecordFromData):
(WebKit::NetworkCache::Storage::dispatchWriteOperation):
(WebKit::NetworkCache::Storage::finishWriteOperationActivity):
(WebKit::NetworkCache::Storage::store):
(WebKit::NetworkCache::Storage::WriteOperation::invokeCompletionHandler):
Deleted.
* Source/WebKit/NetworkProcess/cache/NetworkCacheStorage.h:
(WebKit::NetworkCache::Storage::store): Deleted.
Canonical link: https://commits.webkit.org/292576@main
To unsubscribe from these emails, change your notification settings at
https://github.com/WebKit/WebKit/settings/notifications
_______________________________________________
webkit-changes mailing list
webkit-changes@lists.webkit.org
https://lists.webkit.org/mailman/listinfo/webkit-changes
