feat(routing): Phase-2 reship-monitor + durable Phase-3 trigger (slice 4) #24
+390
−8
Conversation
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
reship_stats(summary): pure summary of what a routed run re-shipped. The load-bearing signal is the SHARD reship ratio (shards shipped / shards present) — appending new big files ships new SOLOS and leaves shards untouched (~0), so healthy GROWTH never trips it; re-shipping many whole SHARDS does, which is small-file churn = the uniform-medium-churn signature Phase 3 (leveling) is for. byte_ratio reported for context only (conflates growth w/ churn, not the trigger). reship_summary_line for the print+email. __main__ size-routing completion: prints the reship line; when shard_ratio > 25% (RESHIP_WATCH_RATIO) flags [RESHIP-WATCH] and the completion email subject/body carry it (warn-only; the operator decides whether to consider Phase 3 for that target). Tests test_reship_monitor.py (7): append -> no watch, wide shard churn -> watch, strict 25% threshold, all-solos never watches, byte-ratio math, empty target, summary line. Suite 238 -> 245. Based on main, not stacked. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…e Phase-3 trigger) Slice 4's [RESHIP-WATCH] email line is a fleeting per-run signal: it fires once routing is enabled, cries wolf on a one-off re-processing pass, and rides in a routine email the operator tunes out. Make it a DURABLE, de-noised, queryable trigger instead. - update_reship_watch(prior_watch, stats, run_epoch, sustain=3): pure fold of a run's reship_stats into a streak block persisted in the target's manifest — consecutive_trips (+1 on a tripped run, reset to 0 on a quiet one), sustained (>= sustain runs), sustained_since, last_ratio/last_run/shards_*, and an `alerted` idempotency latch OWNED BY the deployment alerter (engine only preserves it across trips, clears it on reset so a re-crossing re-alerts once). - route_and_ship computes the watch over the final post-prune object set and always writes the manifest once at the end (records a streak RESET even on an all-skipped run). Adds reship_watch to the return dict. - __main__ surfaces the streak (N/sustain, SUSTAINED?) in the completion print + email so a single trip reads as "not yet sustained", not an action item. The Phase-3 alert (a Teams @mention, fired once on crossing into sustained) is deployment glue that reads this manifest field — kept out of the engine. Tests: 9 pure-function cases (mock summary dicts + prior blocks: increment, reset, sustain-after-N, alerted latch preserve/clear, custom sustain, end-to-end from reship_stats) + 3 persistence cases in route_and_ship (persist+reset, append-does-not-accumulate, repeated-churn-sustains). Suite 245 -> 257. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Jarrod S Doucette (jdoucett)
changed the title
…nag cadence) The durable watch carries two alerter-owned latch fields, alerted + alerted_at (when the deployment alerter last @mentioned). The engine treats both as opaque: preserve across tripped runs, clear on streak reset. This lets the alerter re-nag on a low cadence while a watch stays sustained without the timer being erased by an intervening routed run. Suite 257 -> 258. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign in
to join this conversation on GitHub.
1 participant
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.
Phase 2, Slice 4 — reship-monitor + a DEPENDABLE Phase-3 trigger
A warn-only monitor over each routed run, plus the durable watch state that turns its
signal into a dependable, exception-based Phase-3 trigger (not a fleeting email line).
Monitor (the per-run signal)
The load-bearing signal is the shard reship ratio (shards shipped / shards present):
appending new big files ships new solos and leaves shards untouched (~0), so healthy
growth never trips it; re-shipping many whole shards does — the uniform-medium-churn
signature Phase 3 (leveling) exists for.
byte_ratiois context only.reship_stats(summary)— pure overroute_and_ship's return; shard/byte ratios +watch(True iff shards exist and shard_ratio >
RESHIP_WATCH_RATIO= 25%).Durable watch (the dependable trigger)
A single tripped run is a one-off (a manual re-processing pass cries wolf), and a routine
completion email gets tuned out. So the signal is persisted + de-noised in the target's
manifest:
update_reship_watch(prior_watch, stats, run_epoch, sustain=3)— pure fold into a streak:consecutive_trips(+1 on a tripped run, reset to 0 on a quiet one),sustained(>=
sustainconsecutive trips),sustained_since,last_ratio/last_run, and analertedidempotency latch owned by the deployment alerter (engine only preserves itacross trips and clears it on reset, so a re-crossing re-alerts once).
route_and_shipcomputes the watch over the final post-prune object set and always writesthe manifest once at the end — so an all-skipped run correctly resets the streak. The
watch block is added to the return dict + the completion print/email (a single trip reads
"not yet sustained", not an action item).
The actual exception alert — one Teams @mention to the operator on crossing into a
sustained watch — is deployment glue that reads
manifest.reship_watch; it is kept outof this generic engine (lands in florasense-tools + the Negishi RUNBOOK).
Tests
tests/test_reship_monitor.py(7 monitor + 9 watch-streak) + 3 persistence cases intests/test_routing_wiring.py(persist+reset, append-does-not-accumulate,repeated-churn-sustains). Suite 238 → 257. Based on
main, not stacked.After this: slice 5 (full restore capstone) is the last Phase-2 piece.
🤖 Generated with Claude Code