docs(triage): the line's wording dates a sample against a merged fix — the two routing rows by rbuergi · Pull Request #5190 · Systemorph/MeshWeaver

rbuergi · 2026-09-22T06:28:37Z

Docs only. Conserves the work product of triaging the five newest core bugs (#5176, #5177, #5146, #5175, #2886) against the two routing fixes that merged overnight and are not yet rolled (#5151 for #5009, #5174 for #2299).

What it adds

Beside the existing image-attribution idiom in LogWatchTriage ("read the LINE for the image"), the two rows that the routing cluster produced — so the next reader of a RoutingGrain incident can date every sample against the fix from the sentence it prints, before reading Ops/Status and long before trusting the counter:

site	pre-fix wording	post-fix wording
`[ROUTE] Routing back-pressure …`	`stream destinations queued N, deepest per-destination queue D`	`ordered channels queued N over M stream destination(s), deepest per-channel queue D`
`[ROUTE] Directed delivery to pod hub … failed` on a `DeactivateOnIdle` rejection	`fail:` + `Failed`	`info:` + `ShuttingDown`

Plus the two traps the cluster produced: a closing keyword parses per issue number (PR #5174's Closes #2299's classification half closed #2299 against its own body's statement that the bounce half stays open — reopened, sev:H, now carrying #5177's evidence), and one episode can hold two fingerprints (#5146 / #5177: one address, one window, one set of senders).

What this measured (the dispositions this page now explains)

[Duplicate of #5009] RoutingGrain dispatch budget saturated by head-of-line blocking on cache-destination legs across the portal fleet #5176 → duplicate of RoutingGrain dispatch slots saturate with legs head-of-line blocked behind one shared-cache destination while the second pod starves on load #5009: all 93 folded occurrences print the pre-fix(routing): the ordering FIFO's channel is (destination, stream), not the destination #5151 sentence; both portals run commits (026442fff0, 8a52945608) of which 78bbf49e03 is not an ancestor (control: the memex commit IS an ancestor of the fix merge, i.e. older).
RoutingGrain directed delivery targets a deactivated pod-hub activation instead of re-resolving the pod hub #5177, RoutingGrain fails directed delivery to a transient pod hub instead of retrying or waiting for activation #5146 → duplicates of RoutingGrain turns a stale pod-hub route into a sender-visible DeliveryFailure instead of re-resolving it #2299 (reopened): every sample prints fail:/Failed; 6b3fda2a48 is not an ancestor of either running commit.
#5175 → transferred to MeshWeaver.Plugins#2262 (ThreadSupervisor has 0 files in core); likely downstream of RoutingGrain turns a stale pod-hub route into a sender-visible DeliveryFailure instead of re-resolving it #2299's terminal verdict, plus a Plugins-side never-reset retry budget visible in the samples.
[Duplicate of #1122] sync hub BuildupAction hangs past the 120s startup bound, leaving the hub in FAILED state #2886 → finding on the ticket: two equal 120 s bounds nest and the outer, uninformative one always wins, so the line cannot name the hung action. A separate PR follows.

Verification

step	signal
`dotnet build test/MeshWeaver.Documentation.Test -c Release -warnaserror`	`0 Warning(s)`, `0 Error(s)`, `Time Elapsed 00:00:24.40`
`DocumentationLinkIntegrityTest`	`total: 1, failed: 0, succeeded: 1, skipped: 0`
negative control — one new link deliberately broken	`total: 1, failed: 1, succeeded: 0`, log names the broken target; restored to 0 occurrences

No What's New file (policy whatsnew-cadence). No hard-coded decision dates or names; the dates present are measurements. No addresses to recycle: doc content only.

Mirror-sync: none — no catalog key touched.
Pairs-with: none — no public surface touched.

🤖 Generated with Claude Code

…— the two routing rows A rising counter on an auto-filed incident is not new evidence while the fix is merged and not rolled, and the strongest test of each sample is the log line's own wording, not its timestamp: a timestamp says "after the merge", the wording says which code emitted it. Adds the two rows the routing cluster produced — the back-pressure line's field rename (per-destination → per-channel) and the pod-hub leg's level+word flip (fail:/Failed → info:/ShuttingDown on a DeactivateOnIdle rejection) — beside the existing image-attribution idiom, plus the two traps the cluster produced: a closing keyword parses per issue number, and one episode can hold two fingerprints. DocumentationLinkIntegrityTest: passed 1 / failed 0 — and failed 1 with a link deliberately broken, so the sweep demonstrably reads the page. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Copilot

Copilot review overview

🟡 Changes recommended

Review comments identify unresolved inaccuracies and overstatements in the documentation.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

Clarify identity includes call site and masked message · New

What changed in this PR

Docs-only update adding routing log triage guidance, measured examples, and fingerprinting/issue-closing pitfalls.

Changes:

Documents pre-/post-fix routing log wording.
Adds guidance for dating samples against fixes.
Explains issue-closing and multi-fingerprint traps.

File	Summary
`src/MeshWeaver.Documentation/Data/Architecture/LogWatchTriage.md`	Adds routing-specific triage procedures and examples.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

rbuergi · 2026-09-22T06:33:56Z

+- **The same episode can hold two fingerprints.** The current identity keys on the exception type,
+  so one log site fills one fingerprint when the exception is captured and another when it is not;


Agreed — the identity is hash(WHERE, WHAT, WHICH), and the exception type is one input among the call site and the masked message, not the key. Rewording to say that capturing the same site with and without the exception changes the inputs, so one episode can land on two fingerprints; the split is a property of the inputs differing, not of the type alone. Pushing the reword now.

github-actions · 2026-09-22T06:34:34Z