On 2.12.2013, at 20.50, Mike Abbott <michael.abb...@apple.com> wrote:
>> how [FTS indexing] could be improved for everyone in future > > For sites which set client_limit > 1 it would help performance not to stall > for INDEXER_WAIT_MSECS when polling the indexer for input. Currently dovecot > unwinds back out to the main command loop repeatedly to allow other clients > to use the process but it also stalls the whole process for > INDEXER_WAIT_MSECS every time it finds no input from the indexer, which hurts > responsiveness for those other clients. This can be avoided by removing the > client's I/O from the main ioloop and adding the indexer's instead, or > perhaps by leveraging CLIENT_COMMAND_STATE_WAIT_EXTERNAL. Gets a bit tricky to implement, at least without changing the lib-storage API. I did have some plans for this earlier where lib-storage could call some callback when there is more data available for search/fetch/mailbox_open/etc functions. Currently I’m thinking that most of the reasons for client_limit>1 can be avoided just by moving IMAP IDLE connections to a separate imap-idle process where they wait until they have more work to do. Do you think that would work for you also?
