wbtiger
/
foundationdb
mirror of https://github.com/apple/foundationdb.git
0
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Lower bound version of CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE (#11840)

This commit is contained in:
Syed Paymaan Raza 2024-12-18 23:33:01 -08:00 committed by GitHub
parent a14a0d0fda
commit c2862b8728
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
fdbclient/ServerKnobs.cpp
View File

@ -799,6 +799,7 @@ void ServerKnobs::initialize(Randomize randomize, ClientKnobs* clientKnobs, IsSi
init( CC_MIN_DEGRADATION_INTERVAL, 120.0 );
init( ENCRYPT_KEY_PROXY_FAILURE_TIME, 0.1 ); if ( isSimulated ) ENCRYPT_KEY_PROXY_FAILURE_TIME = 1.0 + deterministicRandom()->random01();
init( CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE, 3 );
init( CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE_MIN, 1 );
init( CC_MAX_EXCLUSION_DUE_TO_HEALTH, 2 );
init( CC_HEALTH_TRIGGER_RECOVERY, false, Atomic::NO );
init( CC_TRACKING_HEALTH_RECOVERY_INTERVAL, 3600.0 );

2
fdbclient/include/fdbclient/ServerKnobs.h
View File

@ -766,6 +766,8 @@ public:
int CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE; // The maximum number of degraded peers when excluding a server. When the
// number of degraded peers is more than this value, we will not exclude
// this server since it may because of server overload.
int CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE_MIN; // Similar to CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE which is an upper
// bound, this is a lower bound.
int CC_MAX_EXCLUSION_DUE_TO_HEALTH; // The max number of degraded servers to exclude by Cluster Controller due to
// degraded health.
bool CC_HEALTH_TRIGGER_RECOVERY; // If true, cluster controller will kill the master to trigger recovery when

3
fdbserver/include/fdbserver/ClusterController.actor.h
View File

@ -3208,7 +3208,8 @@ public:
// For degraded server that are complained by more than SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE, we
// don't know if it is a hot server, or the network is bad. We remove from the returned degraded server list.
for (const auto& badServer : currentDegradedServers) {
if (degradedLinkDst2Src[badServer].size() <= SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE) {
if (degradedLinkDst2Src[badServer].size() >= SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE_MIN &&
degradedLinkDst2Src[badServer].size() <= SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE) {
currentDegradationInfo.degradedServers.insert(badServer);
}
}

1
tests/rare/ClogRemoteTLog.toml
View File

@ -23,6 +23,7 @@ peer_latency_degradation_threshold = 1
peer_latency_degradation_percentile = 0.5
peer_latency_check_min_population = 10
cc_max_exclusion_due_to_health = 3
cc_degraded_peer_degree_to_exclude_min = 2
[[test]]
testTitle = "ClogRemoteTLog"