Re: measuring lwlock-related latency spikes

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> I suggest we optimise that by moving the dirty block into shared
> buffers and leaving it as dirty. That way we don't need to write or
> fsync at all and the bgwriter can pick up the pieces. So my earlier
> patch to get the bgwriter to clean the clog would be superfluous.

[ blink... ] I think you forgot to mention the massive restructuring
needed to cause clog to become a normal relation that the bgwriter and
shared buffer manager would know what to do with. This might be a good
long-term approach but it's not going to produce any near-term joy.

I note BTW that many years ago, the transaction log *was* a normal
relation file, and the current clog code descends directly from
realizing that that was a bad idea. If memory serves, the killer
problem was that a standard relation file doesn't support truncation
from the front; but there may have been other issues as well.

regards, tom lane