christoph/fail
1
Fork 0
Code Actions Packages Releases Activity

jobserver: improved job retry

Browse Source 
FIXME: linear complexity not really hurting here, but not pretty

git-svn-id: https://www4.informatik.uni-erlangen.de/i4svn/danceos/trunk/devel/fail@1312 8c4709b5-6ec9-48aa-a5cd-a96041d1645a
This commit is contained in:
hsc
2012-06-06 14:27:23 +00:00
parent 08ec3548af
commit cfa6153b56
2 changed files with 37 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
core/jobserver/JobServer.cc
View File

@ -244,12 +244,12 @@ void CommThread::sendPendingExperimentData(Minion& minion)
#ifndef __puma
boost::unique_lock<boost::mutex> lock(m_CommMutex);
#endif
if((exp = m_js.m_runningJobs.first()) != NULL) { // 2nd priority
// (This simply gets the first running-job.)
// TODO: Improve selection of parameter-set to be resend (the first is not
// necessarily the best...especially when the specific parameter-set
// causes the experiment-client to terminate abnormally -> endless loop!)
// Further ideas: sequential, random, ...? (+ "retry-counter" for each job)
if((exp = m_js.m_runningJobs.pickone()) != NULL) { // 2nd priority
// (This picks one running job.)
// TODO: Improve selection of parameter set to be resent:
// - currently: Linear complexity!
// - pick entry at random?
// - retry counter for each job?
// Implement resend of running-parameter sets to improve campaign speed
// and to prevent result loss due to (unexpected) termination of experiment