From 63d68ffdbdc095a263b6cc0f2ad106e4b9e4ae1d Mon Sep 17 00:00:00 2001
From: Michael Davis <michael.davis@cern.ch>
Date: Wed, 23 Jan 2019 15:28:27 +0100
Subject: [PATCH] [os-failedrequests] Implements cta-admin failedjobs ls -S
---
objectstore/ArchiveQueue.cpp | 6 +++---
scheduler/OStoreDB/OStoreDB.cpp | 18 ++++++++++++++++++
2 files changed, 21 insertions(+), 3 deletions(-)
diff --git a/objectstore/ArchiveQueue.cpp b/objectstore/ArchiveQueue.cpp
index e0ae0c3f8b..a44e8241c5 100644
--- a/objectstore/ArchiveQueue.cpp
+++ b/objectstore/ArchiveQueue.cpp
@@ -564,9 +564,9 @@ auto ArchiveQueue::getCandidateList(uint64_t maxBytes, uint64_t maxFiles, std::s
auto ArchiveQueue::getCandidateSummary() -> CandidateJobList {
checkPayloadReadable();
CandidateJobList ret;
- for(auto & rqsp: m_payload.archivequeueshards()) {
- ret.candidateBytes += rqsp.shardbytescount();
- ret.candidateFiles += rqsp.shardjobscount();
+ for(auto & aqsp: m_payload.archivequeueshards()) {
+ ret.candidateBytes += aqsp.shardbytescount();
+ ret.candidateFiles += aqsp.shardjobscount();
}
return ret;
}
diff --git a/scheduler/OStoreDB/OStoreDB.cpp b/scheduler/OStoreDB/OStoreDB.cpp
index b2fb41a718..2015c684f8 100644
--- a/scheduler/OStoreDB/OStoreDB.cpp
+++ b/scheduler/OStoreDB/OStoreDB.cpp
@@ -704,6 +704,15 @@ SchedulerDatabase::JobsFailedSummary OStoreDB::getArchiveJobsFailedSummary(log::
auto queueList = re.dumpArchiveQueues(QueueType::FailedJobs);
for(auto &aj : queueList) {
ArchiveQueue aq(aj.address, m_objectStore);
+ try {
+ aq.fetchNoLock();
+ } catch (cta::exception::Exception &ex) {
+ log::ScopedParamContainer params (logContext);
+ params.add("queueObject", aj.address)
+ .add("exceptionMessage", ex.getMessageValue());
+ logContext.log(log::WARNING, "In OStoreDB::getArchiveJobsFailedSummary(): failed to lock/fetch an archive queue.");
+ continue;
+ }
auto summary = aq.getCandidateSummary();
ret.totalFiles += summary.candidateFiles;
ret.totalBytes += summary.candidateBytes;
@@ -1144,6 +1153,15 @@ SchedulerDatabase::JobsFailedSummary OStoreDB::getRetrieveJobsFailedSummary(log:
auto queueList = re.dumpRetrieveQueues(QueueType::FailedJobs);
for(auto &rj : queueList) {
RetrieveQueue rq(rj.address, m_objectStore);
+ try {
+ rq.fetchNoLock();
+ } catch (cta::exception::Exception &ex) {
+ log::ScopedParamContainer params (logContext);
+ params.add("queueObject", rj.address)
+ .add("exceptionMessage", ex.getMessageValue());
+ logContext.log(log::WARNING, "In OStoreDB::getRetrieveJobsFailedSummary(): failed to lock/fetch a retrieve queue.");
+ continue;
+ }
auto summary = rq.getCandidateSummary();
ret.totalFiles += summary.candidateFiles;
ret.totalBytes += summary.candidateBytes;
--
GitLab