Open Bukhtawar opened 1 year ago
During shard relocations before a shard can be marked active, the system makes a call unlink call which is too slow as demonstrated below which is pretty common for a large files i.e the kernel doesn't unlock the inode until all the blocks pointers are returned to the free block list.
Wouldn't the unlink
syscall invoked after the shard is marked active? After peer recovery, the shard is marked active on new node, and the old node will start deleting the files.
Listing down the improvements
unlink
call which is too slow as demonstrated below which is pretty common for a large files i.e the kernel doesn't unlock the inode until all the blocks pointers are returned to the free block list. Now theunlink
call is executed on the cluster applier thread. So if either the file is large or disk slow, this can result in node drops as seen ina) Merge
b) Refresh/Flush
c) Translog Replay
d) Searches on a cold cache
Describe the solution you'd like A clear and concise description of what you want to happen.
Describe alternatives you've considered A clear and concise description of any alternative solutions or features you've considered.
Additional context Add any other context or screenshots about the feature request here.