Re: [PATCH 8/9] maintenance: auto-size incremental-repack batch

On 8/6/2020 1:02 PM, Son Luong Ngoc wrote:

Hi Derrick,

quoted

On Aug 6, 2020, at 18:30, Derrick Stolee via GitGitGadget [off-list ref] wrote:

From: Derrick Stolee <redacted>

When repacking during the 'incremental-repack' task, we use the
--batch-size option in 'git multi-pack-index repack'. The initial setting
used --batch-size=0 to repack everything into a single pack-file. This is
not sustainable for a large repository. The amount of work required is
also likely to use too many system resources for a background job.

Update the 'incremental-repack' task by dynamically computing a
--batch-size option based on the current pack-file structure.

The dynamic default size is computed with this idea in mind for a client
repository that was cloned from a very large remote: there is likely one
"big" pack-file that was created at clone time. Thus, do not try
repacking it as it is likely packed efficiently by the server.

Instead, we select the second-largest pack-file, and create a batch size
that is one larger than that pack-file. If there are three or more
pack-files, then this guarantees that at least two will be combined into
a new pack-file.

I have been using this strategy with git-care.sh [1] with large success.
However it worth to note that there are still edge case where I observed that
pack count keep increasing because using '--batch-size=<second-biggest-pack>+1'
did not resulted in any repacking.
In one case, I have observed a local copy went up to 160+ packs without being able
to repack.

I have been considering whether a strategy such as falling back to the '(3rd biggest
pack size) + 1' and 4th and 5th and so on... when midx repack call resulted in no-op,
as that was how I fixed my repo when the edge case happen.

Such strategy would require a way to detect midx repack to signal when no-op happen,
so something like 'git multi-pack-index repack --batch-size=123456 --exit-code' would
be much desirable.

quoted

Of course, this means that the second-largest pack-file size is likely
to grow over time and may eventually surpass the initially-cloned
pack-file. Recall that the pack-file batch is selected in a greedy
manner: the packs are considered from oldest to newest and are selected
if they have size smaller than the batch size until the total selected
size is larger than the batch size. Thus, that oldest "clone" pack will
be first to repack after the new data creates a pack larger than that.

We also want to place some limits on how large these pack-files become,
in order to bound the amount of time spent repacking. A maximum
batch-size of two gigabytes means that large repositories will never be
packed into a single pack-file using this job, but also that repack is
rather expensive. This is a trade-off that is valuable to have if the
maintenance is being run automatically or in the background. Users who
truly want to optimize for space and performance (and are willing to pay
the upfront cost of a full repack) can use the 'gc' task to do so.

Create a test for this two gigabyte limit by creating an EXPENSIVE test
that generates two pack-files of roughly 2.5 gigabytes in size, then
performs an incremental repack. Check that the --batch-size argument in
the subcommand uses the hard-coded maximum.

Helped-by: Chris Torek [off-list ref]
Reported-by: Son Luong Ngoc <redacted>
Signed-off-by: Derrick Stolee <redacted>

Generally, I have found working with '--batch-size' to be a bit unpredictable.
I wonder if we could tweak the behavior somewhat so that its more consistent
to use and test?

Thanks for continuing to test with this model. My experience has been limited
to the --batch-size=2g option on repos that typically have >15gb of commit
and tree data in their pack directory.

One bit of unpredictability that we've seen is that the --batch-size uses
the "expected size" of a pack-file to select if it should be repacked. This
only tracks the objects that are referenced in that pack, so when we have
new packs that were "un-thinned" by duplicating delta-bases, those affect
the expected size of a repack.

Here is a great reason to have this series be split. While part I stabilizes,
we can take the time to re-evaluate this strategy. It might require updating
the multi-pack-index builtin itself.

One thing to think about is to focus on the different possible sizes of a
repository. If the entire pack-directory is small, then we might as well
repack everything. (What should the limit be? 2gb? configurable?) In the
case of a larger directory, should we just use the --batch-size logic with
that limit, or should we create a new option for which packs to repack?

For example, the recent simulation [1] of downloading fetch packs and
running this maintenance did see extra space that could be recovered with
the current logic. However, what if we could specify "repack everything
except the largest pack" exactly? That would do exactly what this task
is _intending_, as long as that resulting pack-file does not exceed the
2gb maximum by too much.

[1] https://lore.kernel.org/git/d50fbb33-9be3-1c48-2277-8bf894df734f@gmail.com/ (local)

I will think more on this. I'm open to alternate strategies.

Thanks,
-Stolee