[analyzer] Support assigning more resources/higher priority to the dart analyzer.

[modulovalue]

On a machine with e.g. an M5 Pro/Max chip, there are 18 cores available. The machine I'm working on has 64 gigs of RAM and I've observed the analyzer to only utilize at most 5 cores and 10 gigs of RAM.

I'd love to be able to configure a preference, that is, a hint to the Dart analyzer, that makes it use RAM and CPU cores more liberally.

Oftentimes analysis takes 10+ seconds to reach a fixpoint and while the performance issues are known and being worked on AFAIK, perhaps it might be possible to alleviate the performance issues for some users by simply adding support for assigning the analyzer more compute and more RAM?

[modulovalue]

Here are two hard coded memory knobs that I found:

sdk/pkg/analysis_server/lib/src/analysis_server.dart

Lines 609 to 615 in c1a6be4

	ByteStore createByteStore(ResourceProvider resourceProvider) {
	const M = 1024 * 1024 /*1 MiB*/;
	const G = 1024 * 1024 * 1024 /*1 GiB*/;
	const fileCacheSize = 4 * G;
	const memoryCacheSize = 256 * M;

@scheglov you've last increased the memory cache size by 2x in 0e9e359 and the file cache size by 4x in 9c84348

Your commit message appears to imply that it made a noticable difference. Can you comment on whether you think it would be reasonable to make those values configurable as a way to improve performance on projects that are larger than flutter (as you appear to have used flutter as the measuring stick)?

[modulovalue]

I've just patched them to 1 GiB for the fileCacheSize and 10 GiB for the memoryCacheSize and this made a HUGE difference! However, this requires a custom build since the analyzer is run AOT now. Making this configurable without having to build the SDK would be very appreciated.

[modulovalue]

This is the FileByteStore timing page with some custom diagnostics.

Notice how the actual memory cache and file cache sizes exceed the default sizes shipped by the SDK by 6x for the memory cache (!) and almost 2x for the file cache.

So this confirms it, the default cache sizes were too small in my case and increasing them did actually make a significant difference in the responsiveness of the analyzer.

Also, the analyzer already supports a dart-sdk/config/settings.json which could be used for adding support for configuring those values by supporting something like the following:

  {
    "server.byteStore.memoryCacheSize": "4GiB",
    "server.byteStore.fileCacheSize": "16GiB"
  }

I've also noticed that this is even more of an issue in auto-research style environments where agents create worktrees within the repository so any active analyzer session picks them up and starts to struggle. I think making these values configurable and exposing some stats in the diagnostics UI is a change worth applying.