On Mon, Jan 25, 2021 at 06:05:42PM -0800, Darrick J. Wong wrote:

On Mon, Jan 25, 2021 at 05:14:22PM +1100, Dave Chinner wrote:

quoted

Hi folks,

I've been thinking a little about the way we write use cache flushes
recently and I was thinking about how we do journal writes and
whether we need to issue as many cache flushes as we currently do.

Pardon my stream of consciousness, as I try to see if I follow what
you're talking about. :)

quoted

RIght now, every journal write is REQ_PREFLUSH | REQ_FUA, except in
two cases:

1. the log and data devices are different, and we've already issued
a cache flush to the data device, and

2. it's the second write of a split iclog (i.e. tail wrapping) and
so the second bio doesn't need another pre-flush to be run before
submission.

For the purposes of my thinking, #1 above is irrelevant - if we are
able to elide pre-flush from most iclog IOs, we can elide most of
the data dev cache flushes when the log is on a separate device.

However, it's #2 that has got me thinking: the pre-flush isn't
necessary if the journal IO has already received a cache flush that
covers the context of the larger journal write. This behaviour has
been with us since, well, since explicit log IO cache flushes were
added almost 2 decades ago by Steve Lord:

IOWs, we only need to flush if we've written something log related since
the last log write; if we haven't, then we don't need to flush whatever
dirty file data might also be lurking in there, because the user has
fsync for that?

*nod*

The journal is only concerned about metadata, and fsync handles the
explicit data integrity requirements for cache flushing.

quoted

commit 95d97c36e5155075ba2eb22b17562cfcc53fcf96
Author: Steve Lord [off-list ref]
Date:   Fri May 24 14:30:21 2002 +0000

    Add support for drive write cache flushing - should the kernel
        have the infrastructure

But things have changed greatly since then, namely delayed logging
and CIL checkpoints. They have greatly changed how we use iclogs for
writing changes to the journal, but the journal IO mechanisms are
largely unchanged.

That is, the pre-flush is ensuring that all the metadata in the LSN
range of the journal that the iclog sync is about overwrite is
already on stable storage. This means that the metadata writeback in
that range of LSNs has been completed and so we can write the
current log tail into this iclog and correctly order the journal IO
against the already-signalled metadata completion by using a pre-IO
cache flush on the journal IO.

Put another way -- if we've already checkpointed into the filesystem all
of the updates contained in an iclog, we don't have to PREFLUSH before
ovewriting that iclog with a new iclog?

I think you've got the idea, but it helps to think of it as the
range of LSNs that the iclog is going to overwrite in the journal.
i.e. if the range of LSNs that is going to be overwritten has
already been flushed to stable storage, we don't need to do it
again...

quoted

From this observation, it seems to me that we only actually need a
pre-flush to guarantee the metadata being overwritten is stable if the
log tail has moved between the last iclog write and this one. i.e. if
the tail hasn't moved, then a previous iclog write issued the cache
flush to make the metadata in the current LSN overwrite range stable
and so this iclog doesn't need to do it again.

...I think the gist of this is that if our log currently contains iclogs
A, B, and C, we have dirty iclogs D, E, and F that we want to write, and
nobody's been adding things to the log, then we don't have to flush
before writing E because we already took care of that when we wrote D on
top of A?

Yes.

quoted

This implies that that we only need a device cache pre-flush if
(iclog->ic_header.h_lsn != iclog->ic_prev->ic_header.h_lsn).  This
alone would greatly reduce the number of pre-flushes we issue during
a long running CIL checkpoint as tail updates don't occur
frequently.

But the CIL and checkpoint transactions allows us to take this
further.  That is, when we start a checkpoint transaction we've
actually reserved all the space in the log that we require to write
the entire checkpoint. The tail *must* already be beyond where the
end of the checkpoint will reach into the journal because the space
reservations guarantee that the reserve head never wraps around past
the current journal tail.

Hence, when we start a new checkpoint, it will already fit into the
dead LSN range between the head and the tail, and so we only need a
single cache pre-flush to guarantee completion to submission
ordering of the metadata and to-be-issued checkpoint IO into that LSN
range.  That means we only need one data device cache flush per CIL
checkpoint, not one per iclog written in the CIL checkpoint.

Hmm, are you saying we only need to flush caches after we've written the
last iclog in that cil checkpoint?  Or maybe not, given what I see later
on...?

Well, not quite. I'm talking about metadata writeback IO ordering vs
journal overwrites. The pre-flush guarantees metadata to journal
ordering, whilst the FUA/post-flush guarantees journal to metadata
writeback order.

The latter is a constraint of unpinning the log items at IO journal
IO completion - before any of them can be written back in place, the
journal entry for those changes must be on stable storage. Hence the
entire checkpoint that spans a given LSN must be fully written and
stablised before the AIL writes back any metadata tagged with that
LSN.

quoted

IOWs, if we running a 32MB CIL (2GB log) and running full
checkpoints every push, we could reduce the number of cache flushes
by a factor of 100 (256kb iclog = 128 journal IOs for a full CIL
push) to 1000 (32kB journal = 1024 journal IOs). That's a
significant, ongoing reduction in cache flushes.

And then I woundered if we could apply the same logic to
post-journal write cache flushes (REQ_FUA) that guarantee that the
journal writes are stable before we allow writeback of the metadata
in that LSN range (i.e. once they are unpinned). Again, we have a
completion to submission ordering requirement here, only this time
it is journal IO completion to metadata IO submission.

I think this seems reasonable...

quoted

IOWs, I think the same observation about the log head and the AIL
writeback mechanism can be made here: we only need to ensure a cache
flush occurs before we start writing back metadata at an LSN higher

"we start writing back metadata" ... you mean xfsaild writing dirty
buffers back into the fs after a transaction commits to the log?

... and unpins the log items and inserts them into the AIL.

Yes.

quoted

than the journal head at the time of the last cache flush. The first
iclog write of last CIL checkpoint will have ensured all
metadata lower than the LSN of the CIL checkpoint is stable, hence
we only need to concern ourselves about metadata at the same LSN as
that checkpoint. checkpoint completion will unpin that metadata, but
we still need a cache flush to guarantee ordering at the stable
storage level.

So assuming the answer to my last question is yes, then ... an iclog
write will flush the cache to ensure anything written (previously) by
the AIL(?) with a lower LSN than the iclog has made it to disk...?

Yes, that's the ordering guarantee that the pre-flush provides.

quoted

Hence we can use an on-demand AIL traversal cache flush to ensure
we have journal-to-metadata ordering. This will be much rarer than
every using FUA for every iclog write, and should be of similar
order of gains to the REQ_PREFLUSH optimisation.

...and in turn if the AIL wants to write a buffer back to the fs, and
that buffer has an LSN higher than whatever the LSN was the last time we
flushed the disk cache, then the AIL needs to induce a cache flush to
stabilize the log before it actually does that writeback?

Yes, that was my original thinking.

But then I realised that we can avoid that by carefully ordering the
commit record for a checkpoint. If we wait for all previous iclogs
that contain checkpoint items to complete, then we can issue the
commit record with REQ_PREFLUSH | REQ_FUA and ensure the entire
checkpoint is stable in the journal when the commit record is made
stable by the FUA write. (see other delayed mail that should have
been delivered by now)

At this point, we don't need the AIL to do anything new at all - all
the device cache synchronisation is done by the journal, and for a
streaming async transaction workload we would only issue 2
PREFLUSH|FUA journal IOs out of the 100-1000 journal IOs that are
required to write out the checkpoint.

quoted

FWIW, because we use checksums to detect complete checkpoints in
the journal now, we don't actually need to use FUA writes to
guarantee they hit stable storage. We don't have a guarantee in what
order they will hit the disk (even with FUA), so the only thing that
the FUA write gains us is that on some hardware it elides the need
for a post-write cache flush. Hence I don't think we need REQ_FUA,
either.

I suspect that the first iclog in a CIL checkpoint should also use
REQ_FUA, if only to ensure that the log is dirtied as quickly as
possible when a new checkpoint starts. The rest of the checkpoint
ordering or cache flushing just doesn't matter until other external
events require explicit ordering.

Yeah, probably. :)

Having both the first iclog and the commit iclog use both
REQ_PREFLUSH | REQ_FUA and none of the others needing it makes
things a lot simpler to implement...

quoted

The only explicit ordering we really have are log forces. As long as
log forces issue a cache flush when they are left pending by CIL
transaction completion, we shouldn't require anything more here. The
situation is similar to the AIL requirement...

In summary, I think we can get rid of almost all iclog REQ_PREFLUSH
and REQ_FUA usage simply by:

	- first iclog write in a CIL checkpoint uses REQ_PREFLUSH

PREFLUSH?  Didn't you say FUA for the first iclog two paragraphs up?

FUA is not directly needed for metadata vs journal IO ordering. For
getting the log dirty fast, yes, but I don't think that's an
absolute requirement.

quoted

	- AIL traversal to an item with the current head LSN
	  triggers a cache flush

I think this confirms the answer to my question about xfsaild inducing
cache flushes.

And not needed anymore :)

quoted

	- SYNC log force triggers a cache flush if required to
	  stabilise the last checkpoint written to the log,
	  otherwise caller flushes caches as necessary (same as
	  current fsync semantics)

<nod>

And none of the existing log force callers need to change behaviour,
either.

quoted

Given that metadata intensive workloads can result in hundreds to
thousands of journal IOs every second, getting rid of the vast
majority of these cache flush operations would be a major
performance win for such workloads...

Of course, this seems so simple now that I've written it, making me
wonder what I've missed.

"Every time I tangle with the log it takes all day and my brain hurts by
the end so I don't tangle with the log" ? :)

And with that, my brain hurts...

Yeah, it helps to consider just the IO ordering constraints and what
cache flushes are required to guarantee the correct behaviour, not
look at the code.. I certainly didn't look at the code - this was
the result of a thought experiment after learning about how much
REQ_FUA hurts lower layers like dm-thin and watching a simple
reflink send a thousand journal IOs a second down to the block
device....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com