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

Fix hotbackup locking. (#10186) 

* Fix dbserver locking and releasing for hotbackup.

* Fix fix.
This commit is contained in:
Max Neunhöffer 2019-10-11 16:38:13 +02:00 committed by KVS85
parent 56561b3854
commit 4a79205894
1 changed files with 46 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

73
arangod/Cluster/ClusterMethods.cpp
View File

@ -3764,27 +3764,42 @@ arangodb::Result lockDBServerTransactions(std::string const& backupId,
}
// Perform the requests
cc->performRequests(requests, lockWait + 1.0, Logger::BACKUP, false, false);
cc->performRequests(requests, lockWait + 5.0, Logger::BACKUP, false, false);
// Now listen to the results:
// Now listen to the results and report the aggregated final result:
arangodb::Result finalRes(TRI_ERROR_NO_ERROR);
auto reportError = [&](int c, std::string const& m) {
if (finalRes.ok()) {
finalRes = arangodb::Result(c, m);
} else {
// If we see at least one TRI_ERROR_LOCAL_LOCK_FAILED it is a failure
// if all errors are TRI_ERROR_LOCK_TIMEOUT, then we report this and
// this will lead to a retry:
if (finalRes.errorNumber() == TRI_ERROR_LOCAL_LOCK_FAILED) {
c = TRI_ERROR_LOCAL_LOCK_FAILED;
}
finalRes = arangodb::Result(c, finalRes.errorMessage() + ", " + m);
}
};
for (auto const& req : requests) {
auto res = req.result;
int commError = handleGeneralCommErrors(&res);
if (commError != TRI_ERROR_NO_ERROR) {
return arangodb::Result(
TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("Communication error locking transactions on ") + req.destination);
reportError(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("Communication error locking transactions on ")
+ req.destination);
continue;
}
TRI_ASSERT(res.answer != nullptr);
auto resBody = res.answer->toVelocyPackBuilderPtrNoUniquenessChecks();
VPackSlice slc = resBody->slice();
if (!slc.isObject() || !slc.hasKey("error") || !slc.get("error").isBoolean()) {
return arangodb::Result(
TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to freeze transactions for hot backup " + backupId +
": " + slc.toJson());
reportError(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to freeze transactions for hot backup " +
backupId + ": " + slc.toJson());
continue;
}
if (slc.get("error").getBoolean()) {
@ -3792,22 +3807,23 @@ arangodb::Result lockDBServerTransactions(std::string const& backupId,
<< "failed to acquire lock from " << req.destination << ": " << slc.toJson();
auto errorNum = slc.get("errorNum").getNumber<int>();
if (errorNum == TRI_ERROR_LOCK_TIMEOUT) {
return arangodb::Result(errorNum, slc.get("errorMessage").copyString());
reportError(errorNum, slc.get("errorMessage").copyString());
continue;
}
return arangodb::Result(
TRI_ERROR_LOCAL_LOCK_FAILED,
reportError(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("lock was denied from ") + req.destination +
" when trying to check for lockId for hot backup " + backupId +
": " + slc.toJson());
" when trying to check for lockId for hot backup " + backupId +
": " + slc.toJson());
continue;
}
if (!slc.hasKey(lockPath) || !slc.get(lockPath).isNumber() ||
!slc.hasKey("result") || !slc.get("result").isObject()) {
return arangodb::Result(
TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to check for lockId for hot backup " + backupId +
": " + slc.toJson());
reportError(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to check for lockId for hot backup " +
backupId + ": " + slc.toJson());
continue;
}
uint64_t lockId = 0;
@ -3817,19 +3833,22 @@ arangodb::Result lockDBServerTransactions(std::string const& backupId,
<< "acquired lock from " << req.destination << " for backupId "
<< backupId << " with lockId " << lockId;
} catch (std::exception const& e) {
return arangodb::Result(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to get lockId for hot backup " + backupId +
": " + slc.toJson() + ", msg: " + e.what());
reportError(TRI_ERROR_LOCAL_LOCK_FAILED,
std::string("invalid response from ") + req.destination +
" when trying to get lockId for hot backup " + backupId +
": " + slc.toJson() + ", msg: " + e.what());
continue;
}
lockedServers.push_back(req.destination.substr(strlen("server:"), std::string::npos));
}
LOG_TOPIC("c1869", DEBUG, Logger::BACKUP)
<< "acquired transaction locks on all db servers";
if (finalRes.ok()) {
LOG_TOPIC("c1869", DEBUG, Logger::BACKUP)
<< "acquired transaction locks on all db servers";
}
return arangodb::Result();
return finalRes;
}
arangodb::Result unlockDBServerTransactions(std::string const& backupId,