In an effort to improve the total execution time of zxfer, one of the bottlenecks that I encountered was the logic within inspect_delete_snap(). The v1.1.7 implementation uses nested loops to determine which destination snapshots to delete. This logic is even executed when -d is not used. On some of my systems with hundreds of snapshots per dataset, this takes several minutes to execute. e.g. if a source and destination contain 1000 snapshots each, then 1000* 1000 iterations of the loop have to be made each time spawning at least 2 grep + 2 cut commands to compare the snapshot names.
I've refactored the function and moved the nested loops to use temp files, and comm to determine snapshots that don't exist in the destination. This implementation has reduced the execute time from minutes to seconds.
Hi Allan,
In an effort to improve the total execution time of
zxfer, one of the bottlenecks that I encountered was the logic withininspect_delete_snap(). Thev1.1.7implementation uses nested loops to determine which destination snapshots to delete. This logic is even executed when-dis not used. On some of my systems with hundreds of snapshots per dataset, this takes several minutes to execute. e.g. if a source and destination contain 1000 snapshots each, then1000* 1000iterations of the loop have to be made each time spawning at least 2grep+ 2cutcommands to compare the snapshot names.I've refactored the function and moved the nested loops to use temp files, and
commto determine snapshots that don't exist in the destination. This implementation has reduced the execute time from minutes to seconds.Here's a possible implementation.
This has only been tested in FreeBSD 14.0
https://github.com/totalAldo/zxfer/blob/9a7b4e1da5305952863ddbe518403b8c8c18521b/zxfer#L1655-L1789