[Issue] VSCode extension - LSP workspace/diagnostic pulling in an infinite loop on idle

[o-l-a-v]

Description

The extension appears to be missing a proper debounce timer or an unchanged result flag for workspace/diagnostic requests, causing VS Code to aggressively pull full diagnostics every 1–2 seconds even when the workspace is completely idle.

When enabling server logs with .vscode/settings.json -> "ryl.trace.server": "messages", with no open YAML files in the editor, Output "ryl Language Server Trace` get's spammed every 1s or so.

Click to expand

[Trace - 12:58:28 PM] Sending request 'initialize - (0)'.
[Trace - 12:58:29 PM] Received response 'initialize - (0)' in 388ms.
[Trace - 12:58:29 PM] Sending notification 'initialized'.
[Trace - 12:58:29 PM] Sending request 'workspace/diagnostic - (1)'.
[Trace - 12:58:29 PM] Sending notification '$/setTrace'.
[Trace - 12:58:29 PM] Received request 'client/registerCapability - (ryl-register-watchers)'.
[Trace - 12:58:29 PM] Sending response 'client/registerCapability - (ryl-register-watchers)'. Processing request took 1ms
[Trace - 12:58:30 PM] Received response 'workspace/diagnostic - (1)' in 894ms.
[Trace - 12:58:32 PM] Sending notification '$/setTrace'.
[Trace - 12:58:32 PM] Sending request 'workspace/diagnostic - (2)'.
[Trace - 12:58:32 PM] Received response 'workspace/diagnostic - (2)' in 467ms.
[Trace - 12:58:33 PM] Sending notification '$/setTrace'.
[Trace - 12:58:34 PM] Sending request 'workspace/diagnostic - (3)'.
[Trace - 12:58:34 PM] Received response 'workspace/diagnostic - (3)' in 205ms.
[Trace - 12:58:36 PM] Sending request 'workspace/diagnostic - (4)'.
[Trace - 12:58:37 PM] Received response 'workspace/diagnostic - (4)' in 913ms.
[Trace - 12:58:39 PM] Sending request 'workspace/diagnostic - (5)'.
[Trace - 12:58:39 PM] Received response 'workspace/diagnostic - (5)' in 168ms.
[Trace - 12:58:40 PM] Sending notification 'textDocument/didOpen'.
[Trace - 12:58:40 PM] Sending notification 'textDocument/didOpen'.
[Trace - 12:58:40 PM] Sending notification 'textDocument/didOpen'.
[Trace - 12:58:41 PM] Sending request 'workspace/diagnostic - (6)'.
[Trace - 12:58:41 PM] Received response 'workspace/diagnostic - (6)' in 228ms.
[Trace - 12:58:43 PM] Sending request 'workspace/diagnostic - (7)'.
[Trace - 12:58:43 PM] Received response 'workspace/diagnostic - (7)' in 113ms.
[Trace - 12:58:45 PM] Sending request 'workspace/diagnostic - (8)'.
[Trace - 12:58:45 PM] Received response 'workspace/diagnostic - (8)' in 128ms.
[Trace - 12:58:47 PM] Sending request 'workspace/diagnostic - (9)'.
[Trace - 12:58:47 PM] Received response 'workspace/diagnostic - (9)' in 210ms.
[Trace - 12:58:49 PM] Sending request 'workspace/diagnostic - (10)'.
[Trace - 12:58:49 PM] Received response 'workspace/diagnostic - (10)' in 114ms.
[Trace - 12:58:51 PM] Sending request 'workspace/diagnostic - (11)'.
[Trace - 12:58:51 PM] Received response 'workspace/diagnostic - (11)' in 129ms.
[Trace - 12:58:53 PM] Sending request 'workspace/diagnostic - (12)'.
[Trace - 12:58:53 PM] Received response 'workspace/diagnostic - (12)' in 125ms.
[Trace - 12:58:55 PM] Sending request 'workspace/diagnostic - (13)'.
[Trace - 12:58:55 PM] Received response 'workspace/diagnostic - (13)' in 112ms.
[Trace - 12:58:57 PM] Sending request 'workspace/diagnostic - (14)'.
[Trace - 12:58:57 PM] Received response 'workspace/diagnostic - (14)' in 152ms.
[Trace - 12:58:58 PM] Sending request 'workspace/diagnostic - (15)'.
[Trace - 12:58:59 PM] Received response 'workspace/diagnostic - (15)' in 134ms.
[Trace - 12:59:01 PM] Sending request 'workspace/diagnostic - (16)'.
[Trace - 12:59:01 PM] Received response 'workspace/diagnostic - (16)' in 106ms.
[Trace - 12:59:03 PM] Sending request 'workspace/diagnostic - (17)'.
[Trace - 12:59:03 PM] Received response 'workspace/diagnostic - (17)' in 182ms.
[Trace - 12:59:04 PM] Sending request 'workspace/diagnostic - (18)'.
[Trace - 12:59:05 PM] Received response 'workspace/diagnostic - (18)' in 118ms.
[Trace - 12:59:07 PM] Sending request 'workspace/diagnostic - (19)'.
[Trace - 12:59:07 PM] Received response 'workspace/diagnostic - (19)' in 124ms.
[Trace - 12:59:08 PM] Sending request 'workspace/diagnostic - (20)'.
[Trace - 12:59:08 PM] Received response 'workspace/diagnostic - (20)' in 112ms.
[Trace - 12:59:10 PM] Sending request 'workspace/diagnostic - (21)'.
[Trace - 12:59:10 PM] Received response 'workspace/diagnostic - (21)' in 126ms.
[Trace - 12:59:12 PM] Sending request 'workspace/diagnostic - (22)'.
[Trace - 12:59:13 PM] Received response 'workspace/diagnostic - (22)' in 111ms.
[Trace - 12:59:14 PM] Sending request 'workspace/diagnostic - (23)'.
[Trace - 12:59:14 PM] Received response 'workspace/diagnostic - (23)' in 111ms.
[Trace - 12:59:16 PM] Sending request 'workspace/diagnostic - (24)'.
[Trace - 12:59:16 PM] Received response 'workspace/diagnostic - (24)' in 129ms.
[Trace - 12:59:18 PM] Sending request 'workspace/diagnostic - (25)'.
[Trace - 12:59:18 PM] Received response 'workspace/diagnostic - (25)' in 119ms.
[Trace - 12:59:20 PM] Sending request 'workspace/diagnostic - (26)'.
[Trace - 12:59:20 PM] Received response 'workspace/diagnostic - (26)' in 135ms.
[Trace - 12:59:22 PM] Sending request 'workspace/diagnostic - (27)'.
[Trace - 12:59:22 PM] Received response 'workspace/diagnostic - (27)' in 119ms.
[Trace - 12:59:24 PM] Sending request 'workspace/diagnostic - (28)'.
[Trace - 12:59:24 PM] Received response 'workspace/diagnostic - (28)' in 148ms.
[Trace - 12:59:26 PM] Sending request 'workspace/diagnostic - (29)'.
[Trace - 12:59:26 PM] Received response 'workspace/diagnostic - (29)' in 112ms.
[Trace - 12:59:28 PM] Sending request 'workspace/diagnostic - (30)'.
[Trace - 12:59:28 PM] Received response 'workspace/diagnostic - (30)' in 184ms.
[Trace - 12:59:30 PM] Sending request 'workspace/diagnostic - (31)'.
[Trace - 12:59:30 PM] Received response 'workspace/diagnostic - (31)' in 149ms.
[Trace - 12:59:32 PM] Sending request 'workspace/diagnostic - (32)'.
[Trace - 12:59:33 PM] Received response 'workspace/diagnostic - (32)' in 125ms.
[Trace - 12:59:34 PM] Sending request 'workspace/diagnostic - (33)'.
[Trace - 12:59:34 PM] Received response 'workspace/diagnostic - (33)' in 123ms.
[Trace - 12:59:36 PM] Sending request 'workspace/diagnostic - (34)'.
[Trace - 12:59:37 PM] Received response 'workspace/diagnostic - (34)' in 128ms.
[Trace - 12:59:38 PM] Sending request 'workspace/diagnostic - (35)'.
[Trace - 12:59:38 PM] Received response 'workspace/diagnostic - (35)' in 121ms.
[Trace - 12:59:40 PM] Sending request 'workspace/diagnostic - (36)'.
[Trace - 12:59:40 PM] Received response 'workspace/diagnostic - (36)' in 130ms.
[Trace - 12:59:42 PM] Sending request 'workspace/diagnostic - (37)'.
[Trace - 12:59:43 PM] Received response 'workspace/diagnostic - (37)' in 118ms.
[Trace - 12:59:44 PM] Sending request 'workspace/diagnostic - (38)'.
[Trace - 12:59:44 PM] Received response 'workspace/diagnostic - (38)' in 119ms.
[Trace - 12:59:46 PM] Sending request 'workspace/diagnostic - (39)'.
[Trace - 12:59:46 PM] Received response 'workspace/diagnostic - (39)' in 174ms.
[Trace - 12:59:48 PM] Sending request 'workspace/diagnostic - (40)'.
[Trace - 12:59:48 PM] Received response 'workspace/diagnostic - (40)' in 125ms.
[Trace - 12:59:50 PM] Sending request 'workspace/diagnostic - (41)'.
[Trace - 12:59:50 PM] Received response 'workspace/diagnostic - (41)' in 121ms.
[Trace - 12:59:52 PM] Sending request 'workspace/diagnostic - (42)'.
[Trace - 12:59:53 PM] Received response 'workspace/diagnostic - (42)' in 109ms.
[Trace - 12:59:54 PM] Sending request 'workspace/diagnostic - (43)'.
[Trace - 12:59:54 PM] Received response 'workspace/diagnostic - (43)' in 127ms.
[Trace - 12:59:56 PM] Sending request 'workspace/diagnostic - (44)'.
[Trace - 12:59:56 PM] Received response 'workspace/diagnostic - (44)' in 119ms.
[Trace - 12:59:58 PM] Sending request 'workspace/diagnostic - (45)'.
[Trace - 12:59:58 PM] Received response 'workspace/diagnostic - (45)' in 130ms.
[Trace - 1:00:00 PM] Sending request 'workspace/diagnostic - (46)'.
[Trace - 1:00:00 PM] Received response 'workspace/diagnostic - (46)' in 133ms.
[Trace - 1:00:02 PM] Sending request 'workspace/diagnostic - (47)'.
[Trace - 1:00:03 PM] Received response 'workspace/diagnostic - (47)' in 133ms.
[Trace - 1:00:04 PM] Sending request 'workspace/diagnostic - (48)'.
[Trace - 1:00:04 PM] Received response 'workspace/diagnostic - (48)' in 137ms.
[Trace - 1:00:06 PM] Sending request 'workspace/diagnostic - (49)'.
[Trace - 1:00:06 PM] Received response 'workspace/diagnostic - (49)' in 138ms.
[Trace - 1:00:08 PM] Sending request 'workspace/diagnostic - (50)'.
[Trace - 1:00:08 PM] Received response 'workspace/diagnostic - (50)' in 152ms.
[Trace - 1:00:10 PM] Sending request 'workspace/diagnostic - (51)'.
[Trace - 1:00:10 PM] Received response 'workspace/diagnostic - (51)' in 191ms.
[Trace - 1:00:12 PM] Sending request 'workspace/diagnostic - (52)'.
[Trace - 1:00:12 PM] Received response 'workspace/diagnostic - (52)' in 105ms.
[Trace - 1:00:14 PM] Sending request 'workspace/diagnostic - (53)'.
[Trace - 1:00:14 PM] Received response 'workspace/diagnostic - (53)' in 143ms.
[Trace - 1:00:16 PM] Sending request 'workspace/diagnostic - (54)'.
[Trace - 1:00:16 PM] Received response 'workspace/diagnostic - (54)' in 114ms.
[Trace - 1:00:18 PM] Sending request 'workspace/diagnostic - (55)'.
[Trace - 1:00:18 PM] Received response 'workspace/diagnostic - (55)' in 143ms.
[Trace - 1:00:20 PM] Sending request 'workspace/diagnostic - (56)'.
[Trace - 1:00:20 PM] Received response 'workspace/diagnostic - (56)' in 118ms.
[Trace - 1:00:22 PM] Sending request 'workspace/diagnostic - (57)'.
[Trace - 1:00:22 PM] Received response 'workspace/diagnostic - (57)' in 126ms.
[Trace - 1:00:24 PM] Sending request 'workspace/diagnostic - (58)'.
[Trace - 1:00:24 PM] Received response 'workspace/diagnostic - (58)' in 144ms.
[Trace - 1:00:26 PM] Sending request 'workspace/diagnostic - (59)'.
[Trace - 1:00:26 PM] Received response 'workspace/diagnostic - (59)' in 117ms.
[Trace - 1:00:28 PM] Sending request 'workspace/diagnostic - (60)'.
[Trace - 1:00:28 PM] Received response 'workspace/diagnostic - (60)' in 110ms.
[Trace - 1:00:29 PM] Sending request 'workspace/diagnostic - (61)'.
[Trace - 1:00:30 PM] Received response 'workspace/diagnostic - (61)' in 162ms.
[Trace - 1:00:32 PM] Sending request 'workspace/diagnostic - (62)'.
[Trace - 1:00:32 PM] Received response 'workspace/diagnostic - (62)' in 124ms.
[Trace - 1:00:33 PM] Sending request 'workspace/diagnostic - (63)'.
[Trace - 1:00:33 PM] Received response 'workspace/diagnostic - (63)' in 134ms.
[Trace - 1:00:33 PM] Sending notification '$/setTrace'.
[Trace - 1:00:35 PM] Sending request 'workspace/diagnostic - (64)'.
[Trace - 1:00:36 PM] Received response 'workspace/diagnostic - (64)' in 119ms.
[Trace - 1:00:38 PM] Sending request 'workspace/diagnostic - (65)'.
[Trace - 1:00:38 PM] Received response 'workspace/diagnostic - (65)' in 109ms.
[Trace - 1:00:39 PM] Sending request 'workspace/diagnostic - (66)'.
[Trace - 1:00:40 PM] Received response 'workspace/diagnostic - (66)' in 147ms.
[Trace - 1:00:42 PM] Sending request 'workspace/diagnostic - (67)'.
[Trace - 1:00:42 PM] Received response 'workspace/diagnostic - (67)' in 117ms.
[Trace - 1:00:43 PM] Sending request 'workspace/diagnostic - (68)'.
[Trace - 1:00:43 PM] Received response 'workspace/diagnostic - (68)' in 115ms.
[Trace - 1:00:45 PM] Sending request 'workspace/diagnostic - (69)'.
[Trace - 1:00:46 PM] Received response 'workspace/diagnostic - (69)' in 107ms.
[Trace - 1:00:48 PM] Sending request 'workspace/diagnostic - (70)'.
[Trace - 1:00:48 PM] Received response 'workspace/diagnostic - (70)' in 114ms.
[Trace - 1:00:49 PM] Sending request 'workspace/diagnostic - (71)'.
[Trace - 1:00:50 PM] Received response 'workspace/diagnostic - (71)' in 190ms.
[Trace - 1:00:52 PM] Sending request 'workspace/diagnostic - (72)'.
[Trace - 1:00:52 PM] Received response 'workspace/diagnostic - (72)' in 194ms.
[Trace - 1:00:53 PM] Sending request 'workspace/diagnostic - (73)'.
[Trace - 1:00:54 PM] Received response 'workspace/diagnostic - (73)' in 133ms.
[Trace - 1:00:56 PM] Sending request 'workspace/diagnostic - (74)'.
[Trace - 1:00:56 PM] Received response 'workspace/diagnostic - (74)' in 105ms.
[Trace - 1:00:58 PM] Sending request 'workspace/diagnostic - (75)'.
[Trace - 1:00:58 PM] Received response 'workspace/diagnostic - (75)' in 104ms.
[Trace - 1:00:59 PM] Sending request 'workspace/diagnostic - (76)'.
[Trace - 1:01:00 PM] Received response 'workspace/diagnostic - (76)' in 173ms.
[Trace - 1:01:02 PM] Sending request 'workspace/diagnostic - (77)'.
[Trace - 1:01:02 PM] Received response 'workspace/diagnostic - (77)' in 116ms.
[Trace - 1:01:03 PM] Sending request 'workspace/diagnostic - (78)'.
[Trace - 1:01:03 PM] Received response 'workspace/diagnostic - (78)' in 132ms.
[Trace - 1:01:06 PM] Sending request 'workspace/diagnostic - (79)'.
[Trace - 1:01:06 PM] Received response 'workspace/diagnostic - (79)' in 111ms.

[o-l-a-v]

This makes ryl.exe run constantly with a lot of CPU in Windows task manager.

20260805-1357-25.9088047.mp4

[owenlamont]

Sorry for neglecting this @o-l-a-v - I'll see what I can do. Let me figure out if the problem is here or in the ryl-vscode repo.

[owenlamont]

Confirmed from the code on both sides. The two-second cadence belongs to the client library and the CPU cost to the server, so the fix belongs in ryl. Your diagnosis of the missing debounce and missing unchanged flag is right on both counts.

vscode-languageclient re-issues workspace/diagnostic a fixed 2000 ms after every response, forever, with no client-side knob, and does so only because ryl server advertises workspaceDiagnostics: true. ryl then answers each pull immediately with a full walk and lint of every YAML file, returning Full reports with result_id: None, so the client can never be told unchanged. The LSP spec expects the opposite: the workspace request "can be long running and is not bound to a specific workspace or document state", so servers hold it open and answer only on change.

Fix options, all server-side

1. Long-poll: hold the request open and respond, streaming partial results, only when a watched file, open buffer or config changes. Removes idle work entirely. Needs a wake signal from didChangeWatchedFiles, didSave and didChange, and must answer RequestCancelled promptly on client cancel or shutdown.
2. Result ids plus unchanged reports keyed on a content hash, so idle pulls are cheap. Still a directory walk every 2 s, but no linting.
3. Stop advertising workspaceDiagnostics. Simplest; the Problems panel would then only cover open files.

Prior art

• ty's server (astral-sh/ruff, crates/ty_server/src/server/api/requests/workspace_diagnostic.rs) is a worked example of option 1. It hashes each file's diagnostics into a resultId, answers unchanged on a match, and when a whole response would be unchanged or empty it does not respond at all: the request is parked on the session as a SuspendedWorkspaceDiagnosticRequest with a revision snapshot, every state-changing notification retries it, and it responds only once something differs. With a partialResultToken it streams chunks flushed at most every 50 ms.
• ruff (crates/ruff_server/src/server.rs), rust-analyzer (crates/rust-analyzer/src/lsp/capabilities.rs) and gopls (gopls/internal/server/general.go) all set workspaceDiagnostics: false and stay with document pull or push, which is option 3. vscode-eslint is document pull only.
• The client loop is known upstream: Workspace diagnostic pull retries continuously even when sending DiagnosticServerCancellationData error microsoft/vscode-languageserver-node#1261 (2023) reports the same two-second re-pull and found DiagnosticServerCancellationData { retriggerRequest: false } did not stop it. The interval is hard-coded in client/src/common/diagnostic.ts.

So options 1 and 3 both have precedent; option 2 on its own still leaves the idle walk. Whichever lands also needs the bundled ryl in the extension bumped from 0.18.1 to ship (owenlamont/ryl-vscode#30).

Where this lives in the code, and a client-free reproduction

• Client: pullWorkspace() in vscode-languageclient 9.0.1 lib/common/diagnostic.js schedules the next pull via setTimeout(..., 2000) after each response and after up to five consecutive errors. ryl-vscode sets no diagnosticPullOptions.
• Server: spawn_workspace_diagnostic in src/lsp/mod.rs starts a thread running workspace_scan over every root; workspace_report builds every item as Full with result_id: None. The server already cancels the previous in-flight scan when a new pull arrives, which bounds the worker count but not the per-pull cost.
• Reproduce without VS Code: run ryl server over stdio, send initialize (advertising textDocument.diagnostic), initialized, then workspace/diagnostic with previousResultIds: [] twice a second apart. Both responses are full reports for every file with no resultId, each taking the scan time visible in your trace.

[owenlamont]

Implementation sketch for option 1, mapping ty's approach onto the pieces src/lsp/mod.rs already has. Not started; parked here so the work can begin from it.

Sketch: long-polling workspace/diagnostic in ryl, after ty

Result ids. workspace_report and document_diagnostic set result_id: None. Give each report a result_id that hashes the diagnostics for that file (ty: Diagnostics::result_id_from_hash). Then workspace_scan compares each file's id against the client's previous_result_ids and emits WorkspaceDocumentDiagnosticReport::Unchanged on a match. This alone is option 2 and is a prerequisite for the rest.

Suspend instead of respond. Today spawn_workspace_diagnostic always answers. Change the worker's completion path: if every item is Unchanged (or the workspace has no diagnostics at all), do not send the response. Park it on the Session as a suspended request holding the RequestId, the original previous_result_ids and a revision counter (ty: SuspendedWorkspaceDiagnosticRequest with snapshot.revision()). Worker already carries id, cancel and handle, so the suspended state can live alongside workers.

Wake on change. Bump the revision and retry the suspended request from the handle_notification arms that mutate state: didOpen, didChange, didClose, and handle_config_change (which today asks pull clients to re-pull via workspace/diagnostic/refresh). A retry that is still all-unchanged suspends again; otherwise it responds. The refresh request becomes unnecessary for clients that are long-polling but stays for clients that are not.

Cancellation and shutdown. $/cancelRequest for a suspended id must answer RequestCancelled immediately, and shutdown must drain any suspended request the same way, or the client hangs. The existing per-scan cancel flag covers in-flight scans; suspended requests need the same treatment.

Streaming. If the client sends partialResultToken, send WorkspaceDiagnosticReportPartialResult chunks as files complete and finish with an empty final result. ty batches flushes to at most one every 50 ms after measuring repaint lag on large projects; workspace_scan already produces items in parallel, so a channel from the par_iter to a flushing thread fits.

Tests to add. tests/lsp_server.rs: a second pull with the first pull's ids yields all-Unchanged; an idle second pull does not get a response until a didChange; cancel of a suspended pull answers RequestCancelled; shutdown with a suspended pull still exits cleanly. The cancelled-pull test already there stays valid.