----------------------------------------------------------------------- WORKLOG TASK -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- TASK...........: Implement a new modifier for flush table CREATION DATE..: Thu, 22 Oct 2009, 12:25 SUPERVISOR.....: Bothorsen IMPLEMENTOR....: COPIES TO......: Bothorsen CATEGORY.......: Server-RawIdeaBin TASK ID........: 57 (http://askmonty.org/worklog/?tid=57) VERSION........: WorkLog-3.4 STATUS.........: Un-Assigned PRIORITY.......: 60 WORKED HOURS...: 0 ESTIMATE.......: 0 (hours remain) ORIG. ESTIMATE.: 0 PROGRESS NOTES: DESCRIPTION: Create a patch to implement a modifier for "flush table" that allows you to only flush a limited number of blocks. One of the difficulties we have is that we would like to persist some things to disk, but flushing an entire table worth of delay_key_write'd indexes takes minutes, during which time the server is so I/O bound that it cannot serve read traffic. If it was possible to spread this flush over time, we would be able to actually do it without removing the server from read production beforehand. Monty answered this question in an email: The right way to do this is probably to add a background thread that flushes blocks in a 'slow' manner. Note that in the Maria storage engine we do already have this, but I wouldn't recommend you to use this yet; This is more a note about what to expect in the future.. ESTIMATED WORK TIME ESTIMATED COMPLETION DATE ----------------------------------------------------------------------- WorkLog (v3.5.9)