| Message ID | 20230222131100.26472-1-jlayton@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | nfs4.0: add a retry loop on NFS4ERR_DELAY to compound function | expand |
Oops. The subject should have read [pynfs PATCH]: On Wed, 2023-02-22 at 08:11 -0500, Jeff Layton wrote: > The latest knfsd server is "courteous" in that it will not revoke a > lease held by an expired client until there is competing access for it. > When there is competing access, it can now return NFS4ERR_DELAY until > the old client is expired. I've seen this happen when running pynfs in > a loop against a server with only 4g of memory. > > The v4.0 compound handler doesn't retry automatically on NFS4ERR_DELAY > like the v4.1 version does. Add support for it using the same timeouts > as the v4.1 compound handler. > > Cc: Dai Ngo <dai.ngo@oracle.com> > Signed-off-by: Jeff Layton <jlayton@kernel.org> > --- > nfs4.0/nfs4lib.py | 21 +++++++++++++++------ > 1 file changed, 15 insertions(+), 6 deletions(-) > > diff --git a/nfs4.0/nfs4lib.py b/nfs4.0/nfs4lib.py > index 9b074f02b91f..eddcd862bc2f 100644 > --- a/nfs4.0/nfs4lib.py > +++ b/nfs4.0/nfs4lib.py > @@ -338,12 +338,21 @@ class NFS4Client(rpc.RPCClient): > un_p = self.nfs4unpacker > p.reset() > p.pack_COMPOUND4args(compoundargs) > - res = self.call(NFSPROC4_COMPOUND, p.get_buffer()) > - un_p.reset(res) > - res = un_p.unpack_COMPOUND4res() > - if SHOW_TRAFFIC: > - print(res) > - un_p.done() > + res = None > + > + # NFS servers can return NFS4ERR_DELAY at any time for any reason. > + # Just delay a second and retry the call again in that event. If > + # it fails after 10 retries then just give up. > + for i in range(1, 10): > + res = self.call(NFSPROC4_COMPOUND, p.get_buffer()) > + un_p.reset(res) > + res = un_p.unpack_COMPOUND4res() > + if SHOW_TRAFFIC: > + print(res) > + un_p.done() > + if res.status != NFS4ERR_DELAY: > + break > + time.sleep(1) > > # Do some error checking >
diff --git a/nfs4.0/nfs4lib.py b/nfs4.0/nfs4lib.py index 9b074f02b91f..eddcd862bc2f 100644 --- a/nfs4.0/nfs4lib.py +++ b/nfs4.0/nfs4lib.py @@ -338,12 +338,21 @@ class NFS4Client(rpc.RPCClient): un_p = self.nfs4unpacker p.reset() p.pack_COMPOUND4args(compoundargs) - res = self.call(NFSPROC4_COMPOUND, p.get_buffer()) - un_p.reset(res) - res = un_p.unpack_COMPOUND4res() - if SHOW_TRAFFIC: - print(res) - un_p.done() + res = None + + # NFS servers can return NFS4ERR_DELAY at any time for any reason. + # Just delay a second and retry the call again in that event. If + # it fails after 10 retries then just give up. + for i in range(1, 10): + res = self.call(NFSPROC4_COMPOUND, p.get_buffer()) + un_p.reset(res) + res = un_p.unpack_COMPOUND4res() + if SHOW_TRAFFIC: + print(res) + un_p.done() + if res.status != NFS4ERR_DELAY: + break + time.sleep(1) # Do some error checking
The latest knfsd server is "courteous" in that it will not revoke a lease held by an expired client until there is competing access for it. When there is competing access, it can now return NFS4ERR_DELAY until the old client is expired. I've seen this happen when running pynfs in a loop against a server with only 4g of memory. The v4.0 compound handler doesn't retry automatically on NFS4ERR_DELAY like the v4.1 version does. Add support for it using the same timeouts as the v4.1 compound handler. Cc: Dai Ngo <dai.ngo@oracle.com> Signed-off-by: Jeff Layton <jlayton@kernel.org> --- nfs4.0/nfs4lib.py | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)