git@vger.kernel.org mailing list mirror (one of many)
 help / color / mirror / code / Atom feed
From: Derrick Stolee <stolee@gmail.com>
To: Taylor Blau <me@ttaylorr.com>
Cc: "Ævar Arnfjörð Bjarmason" <avarab@gmail.com>,
	git@vger.kernel.org, dstolee@microsoft.com, gitster@pobox.com,
	peff@peff.net, szeder.dev@gmail.com
Subject: Re: Making split commit graphs pick up new options (namely --changed-paths)
Date: Tue, 15 Jun 2021 10:21:56 -0400	[thread overview]
Message-ID: <eb5e2df7-25d8-a878-78ba-cca771265978@gmail.com> (raw)
In-Reply-To: <YMOzDGOo/HbqzkPa@nand.local>

On 6/11/2021 3:01 PM, Taylor Blau wrote:
> On Fri, Jun 11, 2021 at 01:47:28PM -0400, Derrick Stolee wrote:
>> On 6/10/2021 8:50 PM, Taylor Blau wrote:
>>> On Fri, Jun 11, 2021 at 01:56:31AM +0200, Ævar Arnfjörð Bjarmason wrote:
>>>> So yeah, maybe we can just unlink() them right away, or another way to
>>>> handle the race is that load_commit_graph_chain() could just try again
>>>> from the beginning in such a case, and presumably picking up the fresh
>>>> just-rewritten chain.
>>>
>>> I'd probably be in favor of the latter.
>>
>> I want to point out that on Windows we cannot successfully unlink()
>> a layer that is currently being read by another Git process. That
>> will not affect server scenarios (to the best of my knowledge) but
>> is important to many end users.
> 
> Right, but isn't this already a problem today? Since the expiration
> window is zero we are already effectively trying to unlink all merged
> layers immediately:
> 
>   - Marking merged commit-graph layers as expired via
>     mark_commit_graphs() by setting their mtime to "now", and then
> 
>   - Immediately removing all layers which have mtime older than an
>     instant later in expire_commit_graphs().

The commit-graph management we built in Scalar and VFS for Git uses
a non-zero expire time to avoid this problem. However, since we will
clean up the ones that fail to unlink() in a future cleanup, we did
not do a similar thing in Git's background maintenance and have not
had any issues.

> (I almost suggested that a race already exists between multiple writers
> that merge multiple layers of the commit-graph, but that race doesn't
> exist because the commit-graph chain is written before other layers are
> marked and expired.)

And the maintenance buitin solves this by using a maintenance.lock
file to avoid concurrent processes operating on the same repo.

> In any case, it seems like the return value from unlink() is
> deliberately ignored in case another process is holding an expired layer
> open when we try to unlink it. So we'll eventually clean up all layers
> that don't belong to the commit-graph-chain, but at the granularity of
> new writes.

Correct. It also requires new data for the write, or the commit-graph
write will leave early and skip the cleanup IIRC.

> (FWIW, I had to re-read 8d84097f96 (commit-graph: expire commit-graph
> files, 2019-06-18) which mentions that a configuration variable would be
> introduced to change the expiration window, but we don't have any such
> configuration option. It also doesn't make any mention of handling this
> problem on Windows, which made me think that the unlink() calls weren't
> checking their return values by accident when in fact it was probably on
> purpose.)

That config option never appeared, probably because ignoring the
unlink() return was sufficient to get around this problem. Thanks
for digging in and making sure I remembered this correctly.

Thanks,
-Stolee

  reply	other threads:[~2021-06-15 14:27 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-06-10 10:40 Making split commit graphs pick up new options (namely --changed-paths) Ævar Arnfjörð Bjarmason
2021-06-10 17:22 ` Taylor Blau
2021-06-10 18:21   ` Derrick Stolee
2021-06-10 23:56   ` Ævar Arnfjörð Bjarmason
2021-06-11  0:50     ` Taylor Blau
2021-06-11 17:47       ` Derrick Stolee
2021-06-11 19:01         ` Taylor Blau
2021-06-15 14:21           ` Derrick Stolee [this message]
2021-06-15 14:35             ` Ævar Arnfjörð Bjarmason
2021-06-16  1:45               ` Junio C Hamano

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=eb5e2df7-25d8-a878-78ba-cca771265978@gmail.com \
    --to=stolee@gmail.com \
    --cc=avarab@gmail.com \
    --cc=dstolee@microsoft.com \
    --cc=git@vger.kernel.org \
    --cc=gitster@pobox.com \
    --cc=me@ttaylorr.com \
    --cc=peff@peff.net \
    --cc=szeder.dev@gmail.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).