docs(ST-qfpqpxzy): worker-quota-probe.sh + uitgebreide pre-flight runbook + architectuurdoc
- scripts/worker-quota-probe.sh: curl-probe die rate-limit-headers parset en { remaining, limit, pct, reset_at_iso } JSON retourneert
- mcp-integration.md: pre-flight sectie uitgebreid met bash-script voorbeeld, retry-strategie (sleep tot reset+5s), bekende beperking
- claude-question-channel.md: M13 quota-gate flow diagram + pre-flight loop uitleg
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
parent
661601e833
commit
bc6936159d
3 changed files with 96 additions and 26 deletions
|
|
@ -75,5 +75,27 @@ weert).
|
|||
Dit patroon (notification-channel via een bestaande pg_notify-stream) is
|
||||
herbruikbaar — zie `docs/patterns/claude-question-channel.md`.
|
||||
|
||||
## Worker quota-gate flow (M13)
|
||||
|
||||
De `worker_heartbeat` MCP-tool gebruikt hetzelfde `scrum4me_changes`-kanaal met een nieuw event-type `worker_heartbeat`:
|
||||
|
||||
```
|
||||
Worker scrum4me-mcp Postgres SSE-route NavBar
|
||||
| | | | |
|
||||
|--worker_heartbeat(15,low)-->| | | |
|
||||
| |--pg_notify('scrum4me_changes', {type:'worker_heartbeat',is_low:true})-->|
|
||||
| | | |--data:{...}---->|
|
||||
| | | | |--stand-by badge
|
||||
```
|
||||
|
||||
**Pre-flight loop** (vóór elke `wait_for_job`):
|
||||
1. `get_worker_settings` → `min_quota_pct`
|
||||
2. `bash scripts/worker-quota-probe.sh` → `{ pct, reset_at_iso }`
|
||||
3. `worker_heartbeat(last_quota_pct, last_quota_check_at, is_low = pct < min_quota_pct)`
|
||||
4. Als `is_low`: log "wachten tot quota-reset om HH:MM", slaap tot `reset_at_iso + 5s`, herhaal stap 2
|
||||
5. Anders: `wait_for_job` aanroepen
|
||||
|
||||
De solo-store houdt `lowQuotaTokenIds: Set<string>` bij. De NavBar toont een stand-by badge wanneer `lowQuotaTokenIds.size >= connectedWorkers && connectedWorkers > 0`.
|
||||
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -79,46 +79,50 @@ Dit blijft gelden als je tussen jobs door commits, branches of pushes hebt gedaa
|
|||
|
||||
## Pre-flight quota-check
|
||||
|
||||
Vóór elke `wait_for_job` kan de worker controleren of er voldoende Claude-quota is. Stappenplan:
|
||||
Vóór elke `wait_for_job` kan de worker controleren of er voldoende Claude-quota is.
|
||||
|
||||
1. Haal de drempel op: `get_worker_settings` → `min_quota_pct` (bijv. 20).
|
||||
2. Probe de Claude API (rate-limit-headers):
|
||||
**Helper-script:** `scripts/worker-quota-probe.sh` doet een minimale API-call (1 token) en geeft JSON terug:
|
||||
|
||||
```bash
|
||||
RESPONSE=$(curl -sI https://api.anthropic.com/v1/messages \
|
||||
-H "x-api-key: $ANTHROPIC_API_KEY" \
|
||||
-H "anthropic-version: 2023-06-01")
|
||||
|
||||
REMAINING=$(echo "$RESPONSE" | grep -i "anthropic-ratelimit-tokens-remaining" | awk '{print $2}' | tr -d '\r')
|
||||
LIMIT=$(echo "$RESPONSE" | grep -i "anthropic-ratelimit-tokens-limit" | awk '{print $2}' | tr -d '\r')
|
||||
RESET_EPOCH=$(echo "$RESPONSE" | grep -i "anthropic-ratelimit-tokens-reset" | awk '{print $2}' | tr -d '\r')
|
||||
|
||||
QUOTA_PCT=$(( REMAINING * 100 / LIMIT ))
|
||||
```json
|
||||
{ "remaining": 180000, "limit": 200000, "pct": 90, "reset_at_iso": "2026-05-06T13:00:00Z" }
|
||||
```
|
||||
|
||||
3. Vergelijk met drempel. Bij low quota: rapporteer en slaap:
|
||||
**Stappenplan:**
|
||||
|
||||
1. Haal drempel op: `get_worker_settings` → `min_quota_pct` (bijv. 20).
|
||||
2. Probe quota:
|
||||
|
||||
```bash
|
||||
if [ "$QUOTA_PCT" -lt "$MIN_QUOTA_PCT" ]; then
|
||||
RESET_MS=$(date -d "@$RESET_EPOCH" +%s%3N 2>/dev/null || echo 0)
|
||||
NOW_MS=$(date +%s%3N)
|
||||
SLEEP_S=$(( (RESET_MS - NOW_MS) / 1000 + 5 ))
|
||||
echo "Wachten tot quota-reset om $(date -d "@$RESET_EPOCH" '+%H:%M')"
|
||||
sleep "$SLEEP_S"
|
||||
fi
|
||||
PROBE=$(bash scripts/worker-quota-probe.sh)
|
||||
PCT=$(echo "$PROBE" | grep -o '"pct":[0-9]*' | cut -d: -f2)
|
||||
RESET_AT=$(echo "$PROBE" | grep -o '"reset_at_iso":"[^"]*"' | cut -d'"' -f4)
|
||||
```
|
||||
|
||||
4. Meld de quota-state via `worker_heartbeat`:
|
||||
3. Vergelijk en meld via `worker_heartbeat`:
|
||||
|
||||
```
|
||||
worker_heartbeat(
|
||||
last_quota_pct=15,
|
||||
last_quota_check_at="2026-05-06T12:00:00Z",
|
||||
is_low=true
|
||||
last_quota_pct=<PCT>,
|
||||
last_quota_check_at=<nu als ISO-8601>,
|
||||
is_low=<PCT < min_quota_pct>
|
||||
)
|
||||
```
|
||||
|
||||
5. Na de sleep: opnieuw controleren (stap 2) voordat `wait_for_job` wordt aangeroepen.
|
||||
4. Bij low quota: retry-strategie — slaap tot reset + 5s:
|
||||
|
||||
```bash
|
||||
if [ "$PCT" -lt "$MIN_QUOTA_PCT" ]; then
|
||||
RESET_S=$(date -d "$RESET_AT" +%s 2>/dev/null || date -j -f "%Y-%m-%dT%H:%M:%SZ" "$RESET_AT" +%s 2>/dev/null || echo 0)
|
||||
NOW_S=$(date +%s)
|
||||
SLEEP_S=$(( RESET_S - NOW_S + 5 ))
|
||||
echo "Wachten tot quota-reset om $(date -d "$RESET_AT" '+%H:%M' 2>/dev/null || echo "$RESET_AT")"
|
||||
[ "$SLEEP_S" -gt 0 ] && sleep "$SLEEP_S"
|
||||
fi
|
||||
```
|
||||
|
||||
5. Na de sleep: herhaal stap 2 (re-probe) voordat `wait_for_job` wordt aangeroepen.
|
||||
|
||||
**Bekende beperking:** elke probe kost ~1 token; bij 12x/uur is de overhead 12 tokens/uur — verwaarloosbaar ten opzichte van taak-gebruik.
|
||||
|
||||
## Prompt
|
||||
|
||||
|
|
|
|||
44
scripts/worker-quota-probe.sh
Executable file
44
scripts/worker-quota-probe.sh
Executable file
|
|
@ -0,0 +1,44 @@
|
|||
#!/usr/bin/env bash
|
||||
# worker-quota-probe.sh — Probe Claude API rate-limit headers en output JSON.
|
||||
#
|
||||
# Vereist: ANTHROPIC_API_KEY env-var.
|
||||
# Output: { "remaining": N, "limit": N, "pct": N, "reset_at_iso": "..." }
|
||||
# Exit 0 bij succes, exit 1 bij fout (geen API key, curl-fout, parse-fout).
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
if [ -z "${ANTHROPIC_API_KEY:-}" ]; then
|
||||
echo '{"error":"ANTHROPIC_API_KEY not set"}' >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
HEADERS_FILE=$(mktemp /tmp/quota-probe-headers.XXXXXX)
|
||||
trap 'rm -f "$HEADERS_FILE"' EXIT
|
||||
|
||||
# Minimale API-aanroep (1 token) om rate-limit-headers te lezen.
|
||||
HTTP_STATUS=$(curl -s -w "%{http_code}" -D "$HEADERS_FILE" -o /dev/null \
|
||||
-X POST https://api.anthropic.com/v1/messages \
|
||||
-H "x-api-key: $ANTHROPIC_API_KEY" \
|
||||
-H "anthropic-version: 2023-06-01" \
|
||||
-H "content-type: application/json" \
|
||||
-d '{"model":"claude-haiku-4-5-20251001","max_tokens":1,"messages":[{"role":"user","content":"."}]}')
|
||||
|
||||
if [ "$HTTP_STATUS" -ge 500 ]; then
|
||||
echo "{\"error\":\"API returned HTTP $HTTP_STATUS\"}" >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Parse headers (case-insensitive grep)
|
||||
REMAINING=$(grep -i 'anthropic-ratelimit-tokens-remaining:' "$HEADERS_FILE" | awk '{print $2}' | tr -d '\r' || true)
|
||||
LIMIT=$(grep -i 'anthropic-ratelimit-tokens-limit:' "$HEADERS_FILE" | awk '{print $2}' | tr -d '\r' || true)
|
||||
RESET=$(grep -i 'anthropic-ratelimit-tokens-reset:' "$HEADERS_FILE" | awk '{print $2}' | tr -d '\r' || true)
|
||||
|
||||
if [ -z "$REMAINING" ] || [ -z "$LIMIT" ] || [ "$LIMIT" -eq 0 ] 2>/dev/null; then
|
||||
echo '{"error":"rate-limit headers not present in response"}' >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
PCT=$(( REMAINING * 100 / LIMIT ))
|
||||
|
||||
printf '{"remaining":%s,"limit":%s,"pct":%s,"reset_at_iso":"%s"}\n' \
|
||||
"$REMAINING" "$LIMIT" "$PCT" "$RESET"
|
||||
Loading…
Add table
Add a link
Reference in a new issue