Quoting Zygo Blaxell <ce3g8...@umail.furryterror.org>:
On Thu, Sep 12, 2019 at 08:26:04PM -0400, General Zed wrote:
Quoting Zygo Blaxell <ce3g8...@umail.furryterror.org>:
You mean: all metadata size is 156 GB on one of your systems. However, you
don't typically have to put ALL metadata in RAM.
You need just some parts needed for defrag operation. So, for defrag, what
you really need is just some large metadata cache present in RAM. I would
say that if such a metadata cache is using 128 MB (for 2 TB disk) to 2 GB
(for 156 GB disk), than the defrag will run sufficiently fast.
You're missing something (metadata requirement for delete?) in those
estimates.
Total metadata size does not affect how much metadata cache you need
to defragment one extent quickly. That number is a product of factors
including input and output and extent size ratio, the ratio of various
metadata item sizes to the metadata page size, and the number of trees you
have to update (number of reflinks + 3 for extent, csum, and free space
trees).
It follows from the above that if you're joining just 2 unshared extents
together, the total metadata required is well under a MB.
If you're defragging a 128MB journal file with 32768 4K extents, it can
create several GB of new metadata and spill out of RAM cache (which is
currently capped at 512MB for assorted reasons). Add reflinks and you
might need more cache, or take a performance hit. Yes, a GB might be
the total size of all your metadata, but if you run defrag on a 128MB
log file you could rewrite all of your filesystem's metadata in a single
transaction (almost...you probably won't need to update the device or
uuid trees).
If you want to pipeline multiple extents per commit to avoid seeking,
you need to multiply the above numbers by the size of the pipeline.
You can also reduce the metadata cache requirement by reducing the output
extent size. A 16MB target extent size requires only 64MB of cache for
the logfile case.
> Don't forget you have to write new checksum and free space tree pages.
> In the worst case, you'll need about 1GB of new metadata pages for each
> 128MB you defrag (though you get to delete 99.5% of them immediately
> after).
Yes, here we are debating some worst-case scenaraio which is actually
imposible in practice due to various reasons.
No, it's quite possible. A log file written slowly on an active
filesystem above a few TB will do that accidentally. Every now and then
I hit that case. It can take several hours to do a logrotate on spinning
arrays because of all the metadata fetches and updates associated with
worst-case file delete. Long enough to watch the delete happen, and
even follow along in the source code.
I guess if I did a proactive defrag every few hours, it might take less
time to do the logrotate, but that would mean spreading out all the
seeky IO load during the day instead of getting it all done at night.
Logrotate does the same job as defrag in this case (replacing a file in
thousands of fragments spread across the disk with a few large fragments
close together), except logrotate gets better compression.
To be more accurate, the example I gave above is the worst case you
can expect from normal user workloads. If I throw in some reflinks
and snapshots, I can make it arbitrarily worse, until the entire disk
is consumed by the metadata update of a single extent defrag.
In fact, I overcomplicated it in my previous answer.
So, we have a 1TB log file "ultralog" split into 256 million 4 KB
extents randomly over the entire disk. We have 512 GB free RAM and 2%
free disk space. The file needs to be defragmented.
We select some (any) consecutive 512 MB of file segments. They are
certainly localized in the metadata, because we are talking about an
ordered b-tree. We write those 512 MB of file extents to another place
on the partition, defragmented (don't update b-tree yet). Then defrag
calculates the fuse (merge) operation on those written extents. Then
it calculates which metadata updates are necessary. Since we have
selected (at the start) consecutive 512 MB of file segments, the
updates to metadata are certainly localazed. The defrag writes out, in
new extents, the required changes to metadata, then updates the super
to commit.
Easy.
