Re: [Cluster-devel] [RFC PATCH 0/2] dirreadahead system call

From: Steven Whitehouse
Date: Tue Jul 29 2014 - 05:36:44 EST


Hi,

On 28/07/14 22:19, Andreas Dilger wrote:
On Jul 28, 2014, at 6:52 AM, Abhijith Das <adas@xxxxxxxxxx> wrote:
OnJuly 26, 2014 12:27:19 AM "Andreas Dilger" <adilger@xxxxxxxxx> wrote:
Is there a time when this doesn't get called to prefetch entries in
readdir() order? It isn't clear to me what benefit there is of returning
the entries to userspace instead of just doing the statahead implicitly
in the kernel?

The Lustre client has had what we call "statahead" for a while,
and similar to regular file readahead it detects the sequential access
pattern for readdir() + stat() in readdir() order (taking into account if
".*"
entries are being processed or not) and starts fetching the inode
attributes asynchronously with a worker thread.
Does this heuristic work well in practice? In the use case we were trying to
address, a Samba server is aware beforehand if it is going to stat all the
inodes in a directory.
Typically this works well for us, because this is done by the Lustre
client, so the statahead is hiding the network latency of the RPCs to
fetch attributes from the server. I imagine the same could be seen with
GFS2. I don't know if this approach would help very much for local
filesystems because the latency is low.
Well GFS2 is more or less a local fs in the case of no contention. There are issues that arise in the contention case, which is why it is very important to ensure that there is as close to 100% chance of using the inodes which are pre-fetched as possible. Thats why we've not yet tried out algorithms that try to guess when the inodes will be required, since one wrong guess can be quite costly in terms of lock bouncing between nodes. However it does seem that the suggestions made in this thread about using ->lookup in the directory as one way to gather information may be a good way to go, since it will only be called for uncached inodes and in addition it would be equally useful to any possible operation which may follow the lookup.

This syscall might be more useful if userspace called readdir() to get
the dirents and then passed the kernel the list of inode numbers
to prefetch before starting on the stat() calls. That way, userspace
could generate an arbitrary list of inodes (e.g. names matching a
regexp) and the kernel doesn't need to guess if every inode is needed.
Were you thinking arbitrary inodes across the filesystem or just a subset
from a directory? Arbitrary inodes may potentially throw up locking issues.
I was thinking about inodes returned from readdir(), but the syscall
would be much more useful if it could handle arbitrary inodes. For
example, if directories are small then it may be more efficient to
aggregate inodes from multiple directories for each prefetch syscall.
I can't really think of any locking issues that could exist with
"arbitrary list of inodes" that couldn't be hit by having a directory
with hard links to the same list of inodes, so this is something that
needs to be handled by the filesystem anyway.

Since this would be an advisory syscall (i.e. it doesn't really
return anything and cannot guarantee that all the inodes will be in
memory), then if the filesystem is having trouble prefetching the
inodes (e.g. invalid inode number(s) or lock ordering or contention
issues) it could always bail out and leave it to stat() to actually
fetch the inodes into memory when accessed.

There is no way it would be sane to keep inodes locked in the kernel
after prefetch, in case the "stat" never arrives, so the best it can
do is cache the inodes in memory (on the client for network fs), and
it is possible cache pressure or lock contention drops them from cache.

There are always going to be race conditions even if limited to a
single directory (e.g. another client modifies the inode after calling
dirreadahead(), but before calling stat()) that need to be handled.

I think there are a lot of benefits that could be had by the generic
syscall, possibly similar to what XFS is doing with the "bulkstat"
interfaces that Dave always mentions. This would be much more so for
cases were you don't want to stat all of the entries in a directory.

Yes, it probably would be possible to do a kind of bulk lookup like that, although it would also be more complicated locking-wise if we allowed arbitrary inode numbers, rather than doing it on a per directory basis. I can see arguments for both approaches though,

Steve.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/