Concurrent append operations
Hello, I am currently evaluating Ceph for our needs and I have a question about the 'object append' feature. I note that the rados core API supports an 'append' operation, and the S3-compatible interface has too. My question is: does Ceph support concurrent append? I would like to use Ceph as a temporary store, a "buffer" if you will, for incoming data from a variety of sources. Each object would hold data for a particular identifier. I'd like to know if two or more different clients can 'append' to the same object, and the data doesn't overwrite each other, and each 'append' is added to the end of the object? Performance wise we'd likely be performing 15-20 thousand writes per second, so we'd be building a pretty big cluster on very fast flash disk. Data would only reside on the system for about an hour at most before being read and deleted. Cheers, David Bell
The core RADOS api will order these on the osd as it receives the operations from clients, and nothing will break if you submit 2 in parallel. I’m less familiar with the S3 interface but I believe appends there will be ordered by the rgw daemon and so will be much slower. Or maybe it works the same as a normal s3 object overwrite and the second one will win and the first one goes into oblivion? Either way it probably won’t fit your needs. -Greg On Mon, Jan 20, 2020 at 3:42 PM David Bell <david.bell@hotjar.com> wrote:
Hello,
I am currently evaluating Ceph for our needs and I have a question about the 'object append' feature. I note that the rados core API supports an 'append' operation, and the S3-compatible interface has too.
My question is: does Ceph support concurrent append? I would like to use Ceph as a temporary store, a "buffer" if you will, for incoming data from a variety of sources. Each object would hold data for a particular identifier. I'd like to know if two or more different clients can 'append' to the same object, and the data doesn't overwrite each other, and each 'append' is added to the end of the object?
Performance wise we'd likely be performing 15-20 thousand writes per second, so we'd be building a pretty big cluster on very fast flash disk. Data would only reside on the system for about an hour at most before being read and deleted.
Cheers,
David Bell _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
participants (2)
-
David Bell
-
Gregory Farnum