[Bug]: parsing kernel esstra data is very slow #39
Description
Mandatory confirmtion
- follow this project's Code of Conduct.
- follow this project's Contributing Guidelines.
- have checked the issue list, and confirm that a similar issue has not been reported before.
Description
When working with the data for a Linux kernel, the 'esstra' utility is quite slow. It takes about 6 minutes for 'esstra show vmlinux'.
for example: 'time esstra show vmlinux' gave a result of 6m37s, or 397 seconds to perform that operation.
Note that this is with output piped to /dev/null, and on a vmlinux image that is in the kernel's filesystem cache.
It is slower (see below) in other circumstances.
The size of the esstra data for my kernel is approx. 155 megabytes (155583207 to be exact)
A direct objcopy of this data from the vmlinux file takes only 14 seconds.
(e.g. time objcopy --dump-section .esstra=vmlinux.esstra-data vmlinux)
This is the same command that is used to extract the raw data from the file in the 'esstra' utility.
The other operations on the data performed for an 'esstra show' command are decode and parse.
I haven't measured internally to see how much time each of these takes, but my guess would be
that the parse operation is the slow one.
Possibly there is a more efficient yaml parser available which might solve this problem.
I thought that 'esstra show -r vmlinux' (showing only the raw, unparsed data) would be faster, but it is not
IMHO the esstra code should be restructured so that only the objcopy (and not the decode and parsed) are
performed when using the '-r' option to just show the raw esstra data.
Version of ESSTRA
v0.4.0-15-geaade66
Environment
Ubuntu 24.04, using esstra on a Linux kernel build.
Steps to reproduce
$ time esstra show -r vmlinux
...
- File: version-timestamp.c
SHA1: 'fbc84274f45d849495c2bff3eb25260e51e9641c'
real 8m7.341s
user 7m41.788s
sys 0m13.693s
Console logs, visual proof, etc.
No response
Expected results
No response
Additional context
No response