Write failed for page X

Question

Hi,
I get multiple
seek: 83669471002624: (1276694809 * 65536) + 0: Invalid argument
queue.data.log: write failed for page 1276694809
queue.data.log: unable to flush page: 1276694809
error messages when closing the db/env and sometimes also while writing to the db. It always occurs when using a queue without DB_APPEND as access method, but I've also seen it when using btree (especially when using it in the same environment as the queue-db). I tested it using 64MB, 1GB, 2GB and 4GB cache, but the results were pretty much the same. The pagesize is set to its maximum (64KB) and each record has a size of ~61440 bytes.
Thanks,
Thomas

Answer 1

The code that used to produce this error looks like this: http://pastebin.com/LDK51NhV
I don't know why, but after writing the small test program for you the error suddenly doesn't occur anymore (both in the test program and the original code), although I haven't made any changes to the code.
The other problem, which stills persists, and I guess might be related to the page write error, is that the write and read throughput decreases quite dramatically even when using DB_HASH, where I'd assume an access rate which is at least somewhat close to O(1), but it increases linear like all the other methods. The only access method where it's not that bad is the DB_QUEUE (without DB_APPEND) where at least the write throughput only decreases slightly over time.
Edit: After a few runs the error started appearing frequently again (I delete the db + env between every run) - might this be some kind of shared memory / mutex / etc. problem?
Thanks,
Thomas
Edited by: user13300886 on Jun 16, 2010 11:18 AM