A tool to reduce the size of heaptrack profiles by extracting specific timeranges from it.
The main purpose is to improve loading speed of heaptrack-gui on extremely
large (GB to TB-sized) profiles.
Usage:
cargo build --release
gunzip -c large-profile.gz | pv | ./target/release/heaptrack-trim --skip-seconds 123 | gzip > small.gz
gunzip,zcatfor.gzprofiles. For other compression formats, something else has to be used.pvis optional and just to show progress. The tool should at least hit 200MiB/s.gzip, again, to re-compress the file.
Now small.gz will contain only the allocations 123 seconds after the profiling started.
heaptrack-trim does not deal with compression at all, and only filters from stdin to stdout.
Usage: heaptrack-trim --skip-seconds <skip-seconds> [--preserve-time] [--buf-size <buf-size>]
cut out irrelevant parts of heaptrack profiles, to reduce file size
Options:
--skip-seconds skip the first N seconds of the profile. required.
--preserve-time do not rewrite timestamps, leaving the scale of graphs in
heaptrack-gui intact. Makes for easier comparison to the
original profile. However, there will be large, ugly gaps in
the graphs where data was removed.
--buf-size how large should the read and write buffers be? defaults to
1e15 bytes
--help display usage information
-
This does not guarantee proportionally smaller file sizes, for example if your profile is 10 seconds long and you skip the first 5 seconds using
5s, then that does not mean the output file is exactly half the size (even when comparing uncompressed files) -
Some data from the skipped seconds is also included for simplicity, but the vast majority should be removed (see source code)
-
heaptrack-trim can only trim from the beginning of a profile. To trim from the end of a profile, it is easy enough to remove lines from the end of the text file such as
head -10000 profile.txt. Although trimming by time would be nice.
MIT