NFS-RGW: unable to read bs=4m

[supriti]

client_trace_4m_nfs.pcap.txt

Ceph 12.2.1/NFS-Ganesha v2.5.2

I am running fio, and for bs=4m, the read fails. In the rgw log, the error says "non-contiguous write position". But from strace and tcpdump, it does not seem to be case. Is it related to this issue: #204?

Attaching the tcp dump. Github does not allow uploading pcap, so adding txt extension

[dang]

My guess is that this is related to a fundamental part of RGW, namely that it (internally) stores all data in 4M RADOS objects, so a write of >4M needs to be split. This may not be properly implemented in librgw, causing the error.

This issues in #203 are fixed in Ceph now, to that shouldn't be it.

@mattbenjamin Any ideas?

[supriti]

Created ceph upstream tracker: http://tracker.ceph.com/issues/22783

[mattbenjamin]

@dang not immediately, but I'll dig at it

[ffilz]

Closing this due to age, feel free to re-open if it's still of interest and you have explored using a currently supported version of Ganesha (V2.8 or V3 at this time).