This commit is contained in:
Joey Hess 2020-10-13 18:57:30 -04:00
parent e726aee7be
commit e4a8b7b26f
No known key found for this signature in database
GPG key ID: DB12DB0FF05F8F38

View file

@ -0,0 +1,43 @@
[[!comment format=mdwn
username="joey"
subject="""comment 9"""
date="2020-10-13T20:31:30Z"
content="""
100k files: 122364k
40k files: 115476k
1k files: 97920k
0 files: 94808k
So, it's using a little bit more for more files, reasonable if something
is scaling up under load.
But where did that use of 90mb for 0 files come from?
My earlier numbers for 8.20200617 were for the debian package of it. I
built that version with a current toolchain:
100k files: 127240k (vs 37188k with old toolchain)
0 files: 105748k
So, much of the increase is due to a change in ghc or the toolchain or
libraries. Maybe its memory manager is preallocating more or a buffer has
started to default to significantly larger?
For that matter, the old build should not have needed 37mb and I'm pretty
sure it used to be more like 2-5mb?
A haskell hello world still only needs a few kb of memory.
Made git-annex's `main = print "hello"` -- 61716k
Made git-annex exit before it runs git ls-files -- 66892k
Made it exit after starting ls-files and cat-file -- 67320k
Made it exit just before running the first action on the first file, but
after seekFilteredKeys did its processing -- 93300k
So this is partly toolchain or compiler caused, but I don't know if the
extra 30mb it grows by that point is because of the toolchain changing,
or another memory leak.
"""]]