Hawkin
/
arangodb
mirror of https://gitee.com/bigwinds/arangodb
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Correct depiction of servers health status

This commit is contained in:
Kaveh Vahedipour 2017-01-20 09:17:04 +01:00
parent 40c6e2b48c
commit bbb45ca397
1 changed files with 13 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
arangod/Agency/Supervision.cpp
View File

@ -103,13 +103,14 @@ void Supervision::upgradeAgency() {
// Check all DB servers, guarded above doChecks // Check all DB servers, guarded above doChecks
std::vector<check_t> Supervision::checkDBServers() { std::vector<check_t> Supervision::checkDBServers() {
std::vector<check_t> ret; std::vector<check_t> ret;
Node::Children const& machinesPlanned = Node::Children const& machinesPlanned =
_snapshot(planDBServersPrefix).children(); _snapshot(planDBServersPrefix).children();
Node::Children const serversRegistered = Node::Children const serversRegistered =
_snapshot(currentServersRegisteredPrefix).children(); _snapshot(currentServersRegisteredPrefix).children();
bool reportPersistent; bool reportPersistent = false;
std::vector<std::string> todelete; std::vector<std::string> todelete;
for (auto const& machine : _snapshot(healthPrefix).children()) { for (auto const& machine : _snapshot(healthPrefix).children()) {
@ -148,6 +149,8 @@ std::vector<check_t> Supervision::checkDBServers() {
good = true; good = true;
} }
reportPersistent = (heartbeatStatus != lastStatus);
query_t report = std::make_shared<Builder>(); query_t report = std::make_shared<Builder>();
report->openArray(); report->openArray();
report->openArray(); report->openArray();
@ -174,9 +177,6 @@ std::vector<check_t> Supervision::checkDBServers() {
if (good) { if (good) {
if (lastStatus != Supervision::HEALTH_STATUS_GOOD) {
reportPersistent = true;
}
report->add( report->add(
"LastHeartbeatAcked", "LastHeartbeatAcked",
VPackValue(timepointToString(std::chrono::system_clock::now()))); VPackValue(timepointToString(std::chrono::system_clock::now())));
@ -210,7 +210,6 @@ std::vector<check_t> Supervision::checkDBServers() {
// for at least grace period // for at least grace period
if (t.count() > _gracePeriod && secondsSinceLeader > _gracePeriod) { if (t.count() > _gracePeriod && secondsSinceLeader > _gracePeriod) {
if (lastStatus == "BAD") { if (lastStatus == "BAD") {
reportPersistent = true;
report->add("Status", VPackValue("FAILED")); report->add("Status", VPackValue("FAILED"));
FailedServer fsj(_snapshot, _agent, std::to_string(_jobId++), FailedServer fsj(_snapshot, _agent, std::to_string(_jobId++),
"supervision", _agencyPrefix, serverID); "supervision", _agencyPrefix, serverID);
@ -257,6 +256,9 @@ std::vector<check_t> Supervision::checkDBServers() {
// Check all coordinators, guarded above doChecks // Check all coordinators, guarded above doChecks
std::vector<check_t> Supervision::checkCoordinators() { std::vector<check_t> Supervision::checkCoordinators() {
bool reportPersistent = false;
std::vector<check_t> ret; std::vector<check_t> ret;
Node::Children const& machinesPlanned = Node::Children const& machinesPlanned =
_snapshot(planCoordinatorsPrefix).children(); _snapshot(planCoordinatorsPrefix).children();
@ -305,6 +307,8 @@ std::vector<check_t> Supervision::checkCoordinators() {
good = true; good = true;
} }
reportPersistent = (heartbeatStatus != lastStatus);
query_t report = std::make_shared<Builder>(); query_t report = std::make_shared<Builder>();
report->openArray(); report->openArray();
report->openArray(); report->openArray();
@ -329,6 +333,7 @@ std::vector<check_t> Supervision::checkCoordinators() {
} }
if (good) { if (good) {
if (goodServerId.empty()) { if (goodServerId.empty()) {
goodServerId = serverID; goodServerId = serverID;
} }
@ -359,6 +364,9 @@ std::vector<check_t> Supervision::checkCoordinators() {
report->close(); report->close();
if (!this->isStopping()) { if (!this->isStopping()) {
_agent->transient(report); _agent->transient(report);
if (reportPersistent) { // STATUS changes should be persisted
_agent->write(report);
}
} }
} }