From: Jeff Hostetler <git@jeffhostetler.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: Anh Le via GitGitGadget <gitgitgadget@gmail.com>,
git@vger.kernel.org, Timothy Jones <timothy@canva.com>,
Jeff Hostetler <jeffhost@microsoft.com>, Anh Le <anh@canva.com>
Subject: Re: [PATCH] index: add trace2 region for clear skip worktree
Date: Wed, 26 Oct 2022 14:29:09 -0400 [thread overview]
Message-ID: <0bb8dc64-aad1-e3c1-66ef-c2e8d6600189@jeffhostetler.com> (raw)
In-Reply-To: <xmqq8rl2lgl3.fsf@gitster.g>
On 10/26/22 12:01 PM, Junio C Hamano wrote:
> Jeff Hostetler <git@jeffhostetler.com> writes:
>
>> In the worst case, we walk the entire index and lstat() for a
>> significant number of skipped-and-not-present files, then near
>> the end of the loop, we find a skipped-but-present directory
>> and have to restart the loop. The second pass will still run
>> the full loop again. Will the second pass actually see any
>> skipped cache-entries? Will it re-lstat() them? Could the
>> `goto restart` just be a `break` or `return`?
>>
>> I haven't had time to look under the hood here, but I was
>> hoping that these two counters would help the series author
>> collect telemetry over many runs and gain more insight into
>> the perf problem.
>
> Without being able to answer these questions, would we be able to
> interpret the numbers reported from these counters?
>
>> Continuing the example from above, if we've already paid the
>> costs to lstat() the 95% sparse files AND THEN near the bottom
>> of the loop we have to do a restart, then we should expect
>> this loop to be doubly slow. It was my hope that this combination
>> of counters would help us understand the variations in perf.
>
> Yes, I understand that double-counting may give some clue to detect
> that, but it just looked roundabout way to do that. Perhaps
> counting the path inspected during the first iteration and the path
> inspected during the second iteration, separately, without the "how
> many times did we repeat?", would give you a better picture? "After
> inspecting 2400 paths, we need to go back and then ended up scanning
> 3000 paths in the flattened index in the second round" would be
> easier to interpret than "We needed flattening, and scanned 5400
> paths in total in the two iterations".
Good point and I was wondering about whether we might see "2 x 95%"
values for path_count in really slow cases. And yes, it would be
better to have `int path_count[2]` and tally each loop pass
independently.
I guess I was looking for a first-order "where is the pain?" knowing
that we could always dig deeper later. That is, is the lstat's or
is it the ensure-full and index rewrite? Or both?
We have other places that do lstat() over the cache-entries and have
added code to spread the loop across n threads. I doubt that is needed
here and I didn't want to lead with it.
>
>> WRT the `intmax_t` vs just `int`: either is fine.
>
> I thought "int" was supposed to be natural machine word, while
> incrementing "intmax_t" is allowed to be much slower than "int".
> Do we expect more than 2 billion paths?
>
You're right. An `int` would be fine here.
Thanks,
Jeff
next prev parent reply other threads:[~2022-10-26 18:30 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-26 0:05 [PATCH] index: add trace2 region for clear skip worktree Anh Le via GitGitGadget
2022-10-26 3:16 ` Junio C Hamano
2022-10-26 14:13 ` Jeff Hostetler
2022-10-26 16:01 ` Junio C Hamano
2022-10-26 18:29 ` Jeff Hostetler [this message]
2022-10-27 0:04 ` Anh Le
2022-10-28 0:46 ` [PATCH v2] " Anh Le via GitGitGadget
2022-10-28 15:49 ` Derrick Stolee
2022-10-28 17:17 ` Junio C Hamano
2022-10-30 23:28 ` Anh Le
2022-10-28 16:50 ` Jeff Hostetler
2022-10-31 0:56 ` [PATCH v3] " Anh Le via GitGitGadget
2022-10-31 22:34 ` Taylor Blau
2022-11-03 23:04 ` [PATCH v4 0/2] " Anh Le via GitGitGadget
2022-11-03 23:05 ` [PATCH v4 1/2] " Anh Le via GitGitGadget
2022-11-03 23:05 ` [PATCH v4 2/2] index: raise a bug if the index is materialised more than once Anh Le via GitGitGadget
2022-11-05 0:29 ` [PATCH v4 0/2] index: add trace2 region for clear skip worktree Taylor Blau
2022-11-07 20:50 ` Derrick Stolee
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=0bb8dc64-aad1-e3c1-66ef-c2e8d6600189@jeffhostetler.com \
--to=git@jeffhostetler.com \
--cc=anh@canva.com \
--cc=git@vger.kernel.org \
--cc=gitgitgadget@gmail.com \
--cc=gitster@pobox.com \
--cc=jeffhost@microsoft.com \
--cc=timothy@canva.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).