On 06/10/2013 10:35 PM, Amos Jeffries wrote:
> On 11/06/2013 9:35 a.m., Alex Rousskov wrote:
>> Since Rock Store rebuild is done in disker processes, there is probably
>> no code to detect that the rebuild is over in the worker process(es)
>> that handle WCCP.
> If there is no feedback, how does the worker identify when it can start
> sending lookups to the disker?
Workers do not use diskers for lookups (they use a shared memory table).
Workers can start doing those lookups (and submit the corresponding I/O
requests to diskers) virtually immediately after startup. Needless to
say, workers will not find much in the shared memory table in the
beginning because only few entries will be loaded, but that is OK.
HTH,
Alex.
Received on Tue Jun 11 2013 - 22:18:24 MDT
This archive was generated by hypermail 2.2.0 : Wed Jun 12 2013 - 12:00:17 MDT