one-db: retire non-essential telemetry fleet-wide per Elazar - drop network/dns/ping probes + dbsize + kpi.checks collection AND all cross-host telemetry shipping (lezama+whey write nowhere remote; supersedes ONEDB-15 repoint branch); KEEP usage polls + context samples on venus only (efficiency-KPI inputs + compact-monitor actuator data, pending Elazar veto on context)
one-db: retire non-essential telemetry fleet-wide per Elazar - drop network/dns/ping probes + dbsize + kpi.checks collection AND all cross-host telemetry shipping (lezama+whey write nowhere remote; supersedes ONEDB-15 repoint branch); KEEP usage polls + context samples on venus only (efficiency-KPI inputs + compact-monitor actuator data, pending Elazar veto on context). CORRECTION 2026-08-23 (bin-venus-cc, measured): usage.polls/usage.alerts/usage.daily were written ENTIRELY from whey (100% whey rows, zero venus rows ever) via ccusage-watch.sh/ccusage-snapshot.sh shelling out to remote pg — never ran on venus at all, contradicting this WI's own 'on venus only' carve-out from day one. Writer stopped whey-side 2026-08-17 17:16 UTC (owner: bin-whey-cc/nw-whey-cc, not a venus incident). Not grants (checked, telemetry_whey_rw still holds INSERT+SELECT). token-usage-collect.sh (venus, sqlite, live, row 23) covers daily/hourly AGGREGATES only — schema confirmed 2 tables (usage_daily, usage_hourly), no per-poll grain. Per-poll sample stream has NO live successor, gone since 08-17 — real gap, not doc drift. Open: revive ccusage-watch.sh on whey (needs ccusage-11.15-compat rewrite, independently broken since before the stop) vs accept aggregate-only going forward — Elazar's call. Transport for downstream readers: kpi.pensanta.com/export.json (venus, live, JSON, ?days=N|?window=Nh params) now serves the aggregate side, credentialed for automation 2026-08-23.
Activity
-
parent=#1586
-
Script side DONE (sh.git abd2f82): deleted testnetwork/dnsprobe-collect/db-size-monitor/venus-kpi-collect/dnsbench; sys-check v2.3 drops check_kpi_failing. ONEDB-15 closed as retire; lezama telemetry.env deleted (nw-lezama), whey pg_hba:136 pulled+reloaded (nw-whey). Timer removals dispatched: nw-whey (testnetwork, dnsprobe-cold, db-size-monitor system), nw-venus (db-size-monitor, venus-kpi-collect remnants). KEEP fleet-wide per PM ruling A: usage+context ship to venus PG from all hosts; lezama spools-and-drops (GCABA egress). Open: kpi-web dashboard fate (reads retired kpi views), telemetry-retention timer scope.
-
script side landed abd2f82; nw timer removals in flight
-
was: (empty)
-
was: one-db: retire non-essential telemetry fleet-wide per Elazar - drop network/dns/ping probes + dbsize + kpi.checks collection AND all cross-host telemetry shipping (lezama+whey write nowhere remote; supersedes ONEDB-15 repoint branch); KEEP usage polls + context samples on venus only (efficiency-KPI inputs + compact-monitor actuator data, pending Elazar veto on context). CORRECTION 2026-08-23 (nw-venus-cc, bin-whey-cc/bin-venus-cc investigation): the 'usage polls' half of this carve-out was never durably scheduled on venus (no systemd unit/timer/cron ever existed for ccusage-watch.sh) and its writer chain (ccusage-watch.sh -> ccusage-snapshot.sh -> pg usage.daily/usage.polls/usage.alerts) died 2026-08-17 17:16 UTC, independently broken since 07-14 (ccusage 11.15.0 dropped the flags it shells out with). Ruling: this carve-out's intent is satisfied by token-usage-collect.sh (sqlite, 15min, live, registry row 23) — do not resurrect ccusage-watch/snapshot. 'context samples' half unaffected (context-monitor.sh, still live).
-
was: one-db: retire non-essential telemetry fleet-wide per Elazar - drop network/dns/ping probes + dbsize + kpi.checks collection AND all cross-host telemetry shipping (lezama+whey write nowhere remote; supersedes ONEDB-15 repoint branch); KEEP usage polls + context samples on venus only (efficiency-KPI inputs + compact-monitor actuator data, pending Elazar veto on context). CORRECTION 2026-08-23 (nw-venus-cc + bin-whey-cc investigation): ccusage-watch.sh's writer chain (-> ccusage-snapshot.sh -> pg usage.daily/usage.polls/usage.alerts) died 2026-08-17 17:16 UTC, never durably scheduled on venus (no systemd unit/timer/cron ever existed), independently broken since 07-14 (ccusage 11.15.0 dropped the flags it shells out with). token-usage-collect.sh (sqlite, live, registry row 23) covers the daily/hourly AGGREGATE half of this carve-out. MEASURED, not assumed: token-usage.sqlite has only usage_daily/usage_hourly tables, no per-poll row. The per-poll sample stream (usage.polls) has NO live successor and has been gone since 08-17 — this is a real data-loss gap against the carve-out, not doc drift. 'context samples' half unaffected (context-monitor.sh, still live). Open: decide whether per-poll granularity is worth reviving (bin-whey-cc's ccusage-watch.sh chain is broken independently of scheduling and would need a ccusage-11.15-compatible rewrite) or whether daily/hourly aggregates satisfy the original intent — Elazar's call.
-
2026-09-24 02:57 bin-whey-cc: WHEY half verified - no netreport/testnetwork/db-size-monitor/kpi-enforce/kpi-web/ccusage-daily-kpi/context-monitor-whey unit in any scope. NOT verified: lezama and venus halves (host-bound, not queried), and Elazar's pending veto on context samples. Staying inProgress.