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

Bug fix 3.3: Fix supervisor thread crash (#4165) 

* port devel branch to 3.3 of supervisor thread death fix
This commit is contained in:
Matthew Von-Maszewski 2017-12-27 16:34:29 -05:00 committed by Max Neunhöffer
parent ef8fcd101c
commit 392ddde251
1 changed files with 170 additions and 153 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
arangod/Agency/Supervision.cpp
View File

@ -405,8 +405,13 @@ std::vector<check_t> Supervision::check(std::string const& type) {
// Do actual monitoring // Do actual monitoring
for (auto const& machine : machinesPlanned) { for (auto const& machine : machinesPlanned) {
std::string lastHeartbeatStatus, lastHeartbeatAcked, lastHeartbeatTime, std::string lastHeartbeatStatus, lastHeartbeatAcked, lastHeartbeatTime,
lastStatus, serverID(machine.first), lastStatus, serverID(machine.first), shortName;
shortName(_snapshot(targetShortID + serverID + "/ShortName").getString());
// short name arrives asynchronous to machine registering, make sure
// it has arrived before trying to use it
if (LEAF == _snapshot(targetShortID + serverID + "/ShortName").type()) {
shortName = _snapshot(targetShortID + serverID + "/ShortName").getString();
// Endpoint // Endpoint
std::string endpoint; std::string endpoint;
@ -518,8 +523,12 @@ std::vector<check_t> Supervision::check(std::string const& type) {
transient(_agent, *tReport); transient(_agent, *tReport);
} }
} }
} else {
LOG_TOPIC(INFO, Logger::SUPERVISION) <<
"Short name for << " << serverID << " not yet available. Skipping health check.";
} // else
} } // for
return ret; return ret;
} }
@ -621,9 +630,17 @@ void Supervision::run() {
// Do nothing unless leader for over 10 seconds // Do nothing unless leader for over 10 seconds
auto secondsSinceLeader = std::chrono::duration<double>( auto secondsSinceLeader = std::chrono::duration<double>(
std::chrono::steady_clock::now() - _agent->leaderSince()).count(); std::chrono::steady_clock::now() - _agent->leaderSince()).count();
// 10 seconds should be plenty of time for all servers to send
// heartbeat status to new leader (heartbeat is once per second)
if (secondsSinceLeader > 10.0) { if (secondsSinceLeader > 10.0) {
try {
doChecks(); doChecks();
} catch (std::exception const& e) {
LOG_TOPIC(ERR, Logger::SUPERVISION) << e.what() << " " << __FILE__ << " " << __LINE__;
} catch (...) {
LOG_TOPIC(ERR, Logger::SUPERVISION) <<
"Supervision::doChecks() generated an uncaught exception.";
}
} }
} }
@ -989,7 +1006,7 @@ void Supervision::getUniqueIds() {
} catch (std::exception const& e) { } catch (std::exception const& e) {
LOG_TOPIC(ERR, Logger::SUPERVISION) LOG_TOPIC(ERR, Logger::SUPERVISION)
<< "Failed to acquire job IDs from agency: " << "Failed to acquire job IDs from agency: "
<< e.what() << __FILE__ << __LINE__; << e.what() << __FILE__ << " " << __LINE__;
} }
} }