feat: Taskdump viewer & expand inline frames in flamegraphs#377
Merged
Conversation
tokio's `taskdump` feature only compiles on Linux (aarch64/x86/x86_64). The unconditional [dependencies] entry was causing macOS CI builds to fail. Move it to the linux-only target section where it already existed.
jlizen
approved these changes
May 8, 2026
Member
jlizen
left a comment
jlizen
left a comment
There was a problem hiding this comment.
New interface looks great, just some small things.
Consider adding unit tests for TaskDumpEvent parsing and the frame expansion logic.
| callframeSymbols: mapToEntries(trace.callframeSymbols), | ||
| threadNames: mapToEntries(trace.threadNames), | ||
| runtimeWorkers: mapToEntries(trace.runtimeWorkers), | ||
| taskDumps: mapToEntries(trace.taskDumps), |
Member
There was a problem hiding this comment.
Bit surprised not to see any changes to steering, i would have expected the schema to change at least?
Np if planned as a follow up.
| self: 0, | ||
| }); | ||
| // Expand inlined frames: an array entry means multiple frames at one address | ||
| const frames = Array.isArray(entry) ? entry : [entry]; |
Member
There was a problem hiding this comment.
Is it just me or is this reversing the ordering of the flamegraphs, to put the outermost deepest? That seems unintuitive if so.
| for (let fi = frames.length - 1; fi >= 0; fi--) { | ||
| const resolved = frames[fi]; | ||
| const key = resolved ? resolved.symbol : addr || "??"; | ||
| const formatted = resolved ? formatFrame(resolved) : formatFrame(addr, callframeSymbols); |
Member
There was a problem hiding this comment.
This seems more fragile than the old way since frames might be [null].
We could do something like:
Array.isArray(entry) ? entry[0] : entry.
| // Check if clicking an idle region with task dumps | ||
| for (const r of taskDetailHitRegions) { | ||
| if (mx >= r.x1 && mx <= r.x2 && my >= r.y1 && my <= r.y2 && r.taskDumps) { | ||
| showTaskDumpStack(r.taskDumps); |
Member
There was a problem hiding this comment.
probably should be in a try/catch similar to showIdleTimeFlamegraph?
| const fgContainer = document.getElementById("fg-container"); | ||
| const fgInstance = FlamegraphRenderer.createFlamegraph(fgContainer); | ||
|
|
||
| function showTaskDumpStack(dumps) { |
Member
There was a problem hiding this comment.
there is a lot of redundant sidebar code between showTaskDumpStack and showIdleTimeFlamegraph, should we consolidate?
Also consider extracting the dump UI logic into a separate module.
| ctx.strokeStyle = "#444"; | ||
|
|
||
| // Cross-hatch pattern for idle periods with task dumps | ||
| if (hasDump) { |
Member
There was a problem hiding this comment.
feels like we'll want this in a helper sooner or later
rcoh
changed the title
1. Fix inlined-frame iteration order in buildFlamegraphTree. Per blazesym,
an array entry in callframeSymbols is [outermost, ..., innermost] — the
real function at the address is at [0] and [i>0] are inlined callees. The
previous implementation iterated N→0, which inverted the call graph in the
flamegraph (inner inlined functions appeared as parents of the outer
function they were inlined into). Now iterates 0→N, matching caller→callee.
Skip nullish slots so sparse arrays (from out-of-order SymbolTableEntry
events) no longer produce phantom (unknown) tree levels.
2. Wrap showTaskDumpStack click in try/catch matching the
showIdleTimeFlamegraph pattern — errors now surface as a toast instead of
silently failing.
3. Extract renderFlamegraphInSidebar({title, subtitle, samples}) helper.
Both showTaskDumpStack and showIdleTimeFlamegraph now delegate to it,
removing ~40 lines of duplicated sidebar setup.
4. Extract drawCrossHatch(x, y, w, h) helper for the diagonal stripe pattern
used on idle periods that have a task dump.
5. Document taskDumps in the ParsedTrace schema
(dial9-trace-loading/SKILL.md) so agent skills can find it. Extend the
schema validator to understand Map<K, [{obj}]> (array-of-objects map
values).
6. Add unit tests in test_trace_analysis.js covering:
- Flamegraph inline frame ordering (catches the bug fixed in #1)
- Sparse inline arrays with undefined slots
- Unresolved address handling
- TaskDumpEvent parsing shape, sort order, task-id integrity
rcoh
added a commit
that referenced
this pull request
May 8, 2026
1. Fix inlined-frame iteration order in buildFlamegraphTree. Per blazesym,
an array entry in callframeSymbols is [outermost, ..., innermost] — the
real function at the address is at [0] and [i>0] are inlined callees. The
previous implementation iterated N→0, which inverted the call graph in the
flamegraph (inner inlined functions appeared as parents of the outer
function they were inlined into). Now iterates 0→N, matching caller→callee.
Skip nullish slots so sparse arrays (from out-of-order SymbolTableEntry
events) no longer produce phantom (unknown) tree levels.
2. Wrap showTaskDumpStack click in try/catch matching the
showIdleTimeFlamegraph pattern — errors now surface as a toast instead of
silently failing.
3. Extract renderFlamegraphInSidebar({title, subtitle, samples}) helper.
Both showTaskDumpStack and showIdleTimeFlamegraph now delegate to it,
removing ~40 lines of duplicated sidebar setup.
4. Extract drawCrossHatch(x, y, w, h) helper for the diagonal stripe pattern
used on idle periods that have a task dump.
5. Document taskDumps in the ParsedTrace schema
(dial9-trace-loading/SKILL.md) so agent skills can find it. Extend the
schema validator to understand Map<K, [{obj}]> (array-of-objects map
values).
6. Add unit tests in test_trace_analysis.js covering:
- Flamegraph inline frame ordering (catches the bug fixed in #1)
- Sparse inline arrays with undefined slots
- Unresolved address handling
- TaskDumpEvent parsing shape, sort order, task-id integrity
noxware
pushed a commit
to noxware/dial9-tokio-telemetry
that referenced
this pull request
May 9, 2026
…#378) * add: TaskDumps UI to viewer * Address PR dial9-rs#377 review feedback 1. Fix inlined-frame iteration order in buildFlamegraphTree. Per blazesym, an array entry in callframeSymbols is [outermost, ..., innermost] — the real function at the address is at [0] and [i>0] are inlined callees. The previous implementation iterated N→0, which inverted the call graph in the flamegraph (inner inlined functions appeared as parents of the outer function they were inlined into). Now iterates 0→N, matching caller→callee. Skip nullish slots so sparse arrays (from out-of-order SymbolTableEntry events) no longer produce phantom (unknown) tree levels. 2. Wrap showTaskDumpStack click in try/catch matching the showIdleTimeFlamegraph pattern — errors now surface as a toast instead of silently failing. 3. Extract renderFlamegraphInSidebar({title, subtitle, samples}) helper. Both showTaskDumpStack and showIdleTimeFlamegraph now delegate to it, removing ~40 lines of duplicated sidebar setup. 4. Extract drawCrossHatch(x, y, w, h) helper for the diagonal stripe pattern used on idle periods that have a task dump. 5. Document taskDumps in the ParsedTrace schema (dial9-trace-loading/SKILL.md) so agent skills can find it. Extend the schema validator to understand Map<K, [{obj}]> (array-of-objects map values). 6. Add unit tests in test_trace_analysis.js covering: - Flamegraph inline frame ordering (catches the bug fixed in dial9-rs#1) - Sparse inline arrays with undefined slots - Unresolved address handling - TaskDumpEvent parsing shape, sort order, task-id integrity
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
I am pretty stoked about this it came out really nice.
when the idle time has a crosshatch, it means we have a task dump for that period
when you click it, we render it as a flamegraph (there can be multiple futures that you are waiting on)
You can also see a task-flamegraph for the entire task