Am I misunderstand read and write quorum? Or am I losing writes?

Erik Søe Sørensen ess at trifork.com
Fri Apr 12 02:30:14 EDT 2013


Is there any chance that the key in question exists in the bucket prior to the test? If so, then you should update the value (read+write) instead of trying to create it as a whole new object. This has to do with the nature of vector clocks.

/Erik [not familiar with the Python client API]



-------- Oprindelig meddelelse --------
Fra: Rob Speer <rob at luminoso.com>
Dato:
Til: riak-users Users <riak-users at lists.basho.com>
Emne: Am I misunderstand read and write quorum? Or am I losing writes?


My understanding is that if you write a document to a cluster of 4 riak servers with w='quorum', then read from that same cluster with r='quorum', that the read should reflect the result of the right.

My understanding is also that, in the Python client (version 5525f4f3, from master), w = r = 'quorum' by default.

And yet, the following check in my code fails occasionally:

riak_obj = bucket.new(docid, doc)
riak_obj.store()
echo_doc = bucket.get(docid).data
for field in doc:
    assert doc[field] == echo_doc[field], (doc, echo_doc)


I thought perhaps my W and R settings were wrong and I was just getting "eventual consistency", so I tried a very slow version where I waited five seconds before reading the updated document. This seems to fail just as often. So it doesn't just seem like I'm reading a write that hasn't happened yet; it seems that the occasional write may _never happen at all_.

Have I misunderstood something? If not, why would this happen, and how can I track it down?
-- Rob



More information about the riak-users mailing list