1
I Use This!
Activity Not Available

Commits : Listings

Analyzed about 1 year ago. based on code collected about 1 year ago.
Jan 18, 2023 — Jan 18, 2024
Commit Message Contributor Files Modified Lines Added Lines Removed Code Location Date
Make -s option to store not be on by deafult (previously it was on if there was replication). We weren't greedy enough on small clusters. Larger clusters might need a more elegant scheduling, but serialization by default probably isn't the asnwer. More... almost 13 years ago
just in case More... almost 13 years ago
Worked out the right way to create the hash XX/YY structure in syncdir and how to handle rm cases that should invalidate it. More... almost 13 years ago
add -e option to map which causes nodes to invalidate cached output that resulted in a non-zero exit status. If the sync lock was from our node, then remove it and run again. If another node holds the lock, then leave it alone. But if the previous holder is gone, nobody will invalidate it. XXX? More... almost 13 years ago
simplify parsing More... almost 13 years ago
I think we weren't aggressive enough with lots of short-lived work items. Sleep less and scale up by 1.x rather than 10x. Also reset polling internval when we reap something. More... almost 13 years ago
Fixed multiple bugs with input caching. 1. Undiscovered bug in continuous jobs 2. After starting all inputs, we would re-enumerate and start again any that hasn't finished. That led to the same work item running multiple times concurrently. More... almost 13 years ago
simplify and fix rsync glob handling More... almost 13 years ago
rework reduce operators (again) to improve syntax and semantics. I think this one works and will stick. More... almost 13 years ago
change FileLocker behavior so that nodes only sync (lock) when they are done instead of when they start. This way a slow node doesn't keep an item from being completed when another node could. This reduces are impact on the NFS file server for sync files. This does allow probabilistically for some duplication of effort, but helps solve the slow-node problem. More... almost 14 years ago
don't call dynamic load updates when !dynamicload More... almost 14 years ago
bugfix: bad function name in obscure error handler More... almost 14 years ago
add fm store -s More... almost 14 years ago
handle possible divide by 0 More... about 14 years ago
report "No processes completed" rather than "No processes ran" --- they could still be running More... about 14 years ago
check for ~/.filemap.conf after ./filemap.conf and before /etc/filemap.conf More... about 14 years ago
more dynamicload improvements More... about 14 years ago
fix replicate bug More... about 14 years ago
put usage in readme More... about 14 years ago
dynamicload bugfix More... about 14 years ago
should fix replicate bug More... about 14 years ago
dynamicload improvements including D state monitoring and better CPU idle monitoring More... about 14 years ago
bugfixes from last patch.
MIke Fisk
as Michael E. Fisk
More... over 14 years ago
Abstract-out (and centralize) global synchronization into a class (that we can replace or have other options for). More... over 14 years ago
document rewriteList semantics More... over 14 years ago
add rewriteList() More... over 14 years ago
used more unique function name for (numcmp)tokenize More... over 14 years ago
trying really hard to get the different reduces working but using a lot of strings now to keep state, maybe thats ok More... over 14 years ago
got the syntax added for the 4 different kinds of reduce we're doing now, the parser works, thinking about using the reduce operator instead of True when making the job More... over 14 years ago
better fix for the sync path of a file More... over 14 years ago