On Fri, Jan 30, 2009 at 1:03 AM, Adam Kocoloski <[email protected]> wrote: > Hi Jeff, it's starting to make some more sense now. How big are the normal > attachments? At present, Couch encodes all attachments using Base64 and > inlines them in the JSON representation of the document during replication. > We'll fix this in the 0.9 release by taking advantage of new support for > multipart requests[1], but until then replicating big attachments is iffy at > best. Regards, > > Adam > > [1] https://issues.apache.org/jira/browse/COUCHDB-163 >
Hi Adam, Of the 282 attachments, 80 or so are 4-8MB, the others are a couple of hundred k to < 4MB, each document has 0-2 attachments so the documents vary from < 1M to 9M in size. There are 188 documents with attachments. If I built the db with just the 88 largest documents and tried to replicate it would work. When replicating the entire test db there seemed to be some point that the remote machine(.52) could not return attachments fast enough that local would not time out waiting on a response. Attempted retries would snowball and the entire process would slow down progressively. The local couch process (.192) would sometimes die completely when it had encountered "too many" timeout/retry events. Although this problem I can't replicate without using the entire test db. 0.9.0a739174-incubating seems to be resilient to this scenario. Although I can't replicate without error, the couch process doesn't go away with the same test set. Thank you for you help in this. -Jeff
