Repository navigation

Health-check heuristic fix, operator drain/block controls, and websocket close-code correctness - #526

Merged

oten91 merged 22 commits into

mainfrom

fix/cometbft-health-check-heuristic

Aug 7, 2026

Commits on Aug 6, 2026
- fix(health-checks): thread the JSON-RPC method into both heuristic sites
  
  oten91
  and
  claude
  committed
  65802ea
- docs(config): catch the schema up to the health-check keys that ship
  
  oten91
  and
  claude
  committed
  a722e32
- feat(admin): per-operator reputation drain endpoint
  
  oten91
  and
  claude
  committed
  933bccf
- feat(metrics): per-connection websocket frame-rate histogram
  
  oten91
  and
  claude
  committed
  77fc889
- docs: reputation drain endpoint + WS frame-rate distribution
  
  oten91
  and
  claude
  committed
  f5419e7
Commits on Aug 7, 2026
- fix(admin): drain survives storage refresh; target by URL not node id
  
  oten91
  and
  claude
  committed
  95e89c0
- feat(admin): fleet-wide reputation drains with a hard expiry ceiling
  
  oten91
  and
  claude
  committed
  234d160
- test(admin): pin drain isolation from the scoring system
  
  oten91
  and
  claude
  committed
  3ef1e3d
- feat(metrics): report admin drains separately from earned cooldowns
  
  oten91
  and
  claude
  committed
  03de910
- fix(admin): key drains by operator, not by resolved endpoint keys
  
  oten91
  and
  claude
  committed
  017bc5c
- fix(admin): apply the drain to the set the filter actually returns
  
  oten91
  and
  claude
  committed
  d79b126
- test(shannon): selection-path harness for routing changes
  
  oten91
  and
  claude
  committed
  502c523
- fix(websocket): close the endpoint connection cleanly on rebind
  
  oten91
  and
  claude
  committed
  57c074e
- test(qos): restore the evm build after the reputation interface grew
  
  oten91
  and
  claude
  committed
  cacce39
- fix(websocket): close health-check probes with a handshake, not a socket drop
  
  oten91
  and
  claude
  committed
  b4640ad
- fix(websocket): close live connections on SIGTERM instead of dropping them
  
  oten91
  and
  claude
  committed
  efef6d4
- fix(websocket): send each peer a close code that fits its role
  
  oten91
  and
  claude
  committed
  1d6a29a
- fix(admin): apply the drain to the endpoint a connection is already bound to
  
  oten91
  and
  claude
  committed
  ca5cfbe
- feat(protocol): ban operator domains from serving an RPC type on every service
  
  oten91
  and
  claude
  committed
  9bde542
- test(protocol): cover the NewProtocol domain-blocklist wiring
  
  oten91
  and
  claude
  committed
  999fbf0
- docs: correct two stale claims about drains and the over-servicing brake
  
  oten91
  and
  claude
  committed
  a842f00
- refactor(test): use placeholder domains in fixtures and examples
  
  oten91
  and
  claude
  committed
  b8c0257

Navigation Menu

Repository navigation

Health-check heuristic fix, operator drain/block controls, and websocket close-code correctness - #526

Health-check heuristic fix, operator drain/block controls, and websocket close-code correctness#526
oten91 merged 22 commits into
mainfrom
fix/cometbft-health-check-heuristic

Commits on Aug 6, 2026

fix(health-checks): thread the JSON-RPC method into both heuristic sites

docs(config): catch the schema up to the health-check keys that ship

feat(admin): per-operator reputation drain endpoint

feat(metrics): per-connection websocket frame-rate histogram

docs: reputation drain endpoint + WS frame-rate distribution

Commits on Aug 7, 2026

fix(admin): drain survives storage refresh; target by URL not node id

feat(admin): fleet-wide reputation drains with a hard expiry ceiling

test(admin): pin drain isolation from the scoring system

feat(metrics): report admin drains separately from earned cooldowns

fix(admin): key drains by operator, not by resolved endpoint keys

fix(admin): apply the drain to the set the filter actually returns

test(shannon): selection-path harness for routing changes

fix(websocket): close the endpoint connection cleanly on rebind

test(qos): restore the evm build after the reputation interface grew

fix(websocket): close health-check probes with a handshake, not a socket drop

fix(websocket): close live connections on SIGTERM instead of dropping them

fix(websocket): send each peer a close code that fits its role

fix(admin): apply the drain to the endpoint a connection is already bound to

feat(protocol): ban operator domains from serving an RPC type on every service

test(protocol): cover the NewProtocol domain-blocklist wiring

docs: correct two stale claims about drains and the over-servicing brake

refactor(test): use placeholder domains in fixtures and examples