Re: pg_prewarm - Mailing list pgsql-hackers

<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> pgfincore does not use the postgresql buffer manager, it uses the posix<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> calls. It can proceed per block or full relation.<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> >
<pstyle=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> Both need POSIX_FADVISE compatible system to be efficient.<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> >
<pstyle=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> The main difference between pgfincore and pg_prewarm about full relation<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> warm is that pgfincore will make very few system calls when pg_prewarm<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> will do much more.<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> <p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
That'sa fair complaint, but I'm not sure it matters in practice,<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> because I think that in
reallife the time spent prewarming is going<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> to be dominated by I/O, not system call
time.Now, that's not an<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> excuse for being less efficient, but I actually did have a
reasonfor<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> doing it this way, which is that it makes it work on systems that<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>don't support POSIX_FADVISE, like Windows and MacOS X. Unless I'm<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
mistakenor it's changed recently, pgfincore makes no effort to be<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> cross-platform, whereas
pg_prewarmshould be usable anywhere that<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> PostgreSQL is, and you'll be able to do
prewarmingin any of those<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> places, though of course it may be a bit less efficient
without<pstyle=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> POSIX_FADVISE, since you'll have to use the "read" or "buffer" mode<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>rather than "prefetch". Still, being able to do it less efficiently<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> is
betterthan not being able to do it at all.<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> <p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
Again,I'm not saying this to knock pgfincore: I see the advantages of<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> its approach in exposing
awhole suite of tools to people running on,<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> well, the operating systems on which the
largestnumber of people run<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> PostgreSQL. But I do think that being cross-platform is an
advantage,<pstyle=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> and I think it's essential for anything we'd consider shipping as a<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>contrib module. I think you could rightly view all of this as<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
pointingto a deficiency in the APIs exposed by core: there's no way<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> for anything above the
smgrlayer to do anything with a range of<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> blocks, which is exactly what we want to
dohere. But I wasn't as<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> interested in fixing that as I was in getting something
whichdid what<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> I needed, which happened to be getting the entirety of a relation into<p
style="margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>shared_buffers without much ado.<p style="-qt-paragraph-type:empty; margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;"><br /><p
style="margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">Agreed,pgfincore first use was to analyze cache usage and performance impacts. (this works with
systemshaving mincore(), not only linux, only windows is really different and while I can add the support for it, I've
neverbeen requested for that, I can do if it helps going to contrib/ if someone care). <p
style="-qt-paragraph-type:empty;margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;"><br /><p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">Warming with pg_prewarm looks
reallycool and it does the job. Pgfincore only advantage here are that if you call POSIX_FADVISE on whole file, the
kernelwill *try* to load as much of possible while not destructing the cache its have. My experience is that if you
callblock-per-block all the blocks you touch are in cache (and eviction can occur more often). <p
style="-qt-paragraph-type:empty;margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;"><br /><p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> <p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> The current implementation of pgfincore allows to make a snapshot and<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> restore via pgfincore or via pg_prewarm (just need some SQL-fu for the<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>> later).<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> <p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> Indeed.<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">><p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> Just to make completely clear my position on pgfincore vs.
pg_prewarm,<pstyle=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> I think they are complementary utilities with a small overlap. I<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>think that the prewarming is important enough to a broad enough group<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> of
peoplethat we should find some way of exposing that functionality<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> in core or contrib, and
Iwrote pg_prewarm as a minimalist<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> implementation of that concept. I am not necessarily
opposedto<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> someone taking the bull by the horns and coming up with a grander<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>vision for what kind of tool we pull into the core distribution -<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
eitherby extending pg_prewarm, recasting pgfincore as a contrib<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> module with appropriate
cross-platformsauce, or coming up with some<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> third approach that is truly the one ring
torule them all and in the<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> darkness bind them. At the same time, I want to get
somethingdone<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> for 9.3 and I don't want to make it harder than it needs to be. I<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>honestly believe that just having an easy way to pull stuff into<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">>
memory/shared_bufferswill give us eighty to ninety percent of what<p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> people need in this
area;we can do more, either in core or elsewhere,<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;">> as the motivation may strike us.<p
style="margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">><p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">> Attached is an updated patch, with fixes for documentation
typonoted<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0;
text-indent:0px;-qt-user-state:0;">> by Jeff Janes and some addition documentation examples also inspired<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">>by comments from Jeff.<p style="-qt-paragraph-type:empty; margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;"><br /><p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">Load-per-blockis indeed very useful as the Slave can really catch-up more quickly with the workload
incase of switchover for example (this is why I've moved pgfincore results in a varbit that can be shared with the
slavesmore easily).<p style="-qt-paragraph-type:empty; margin-top:0px; margin-bottom:0px; margin-left:0px;
margin-right:0px;-qt-block-indent:0; text-indent:0px; -qt-user-state:0;"><br /><p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">I have no
problemdeprecating overlapping features from pgfincore as soon as I can do a «depend:pg_prewarm[os_warm]» :)<p style="
margin-top:0px;margin-bottom:0px; margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">...Itwould have been better to split pgfincore analyze and warming parts times ago, anyway.<p
style="-qt-paragraph-type:empty;margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;"><br /><p style=" margin-top:0px; margin-bottom:0px;
margin-left:0px;margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">-- <p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px; -qt-user-state:0;">Cédric
Villemain+33 (0)6 20 30 22 52<p style=" margin-top:0px; margin-bottom:0px; margin-left:0px; margin-right:0px;
-qt-block-indent:0;text-indent:0px; -qt-user-state:0;">http://2ndQuadrant.fr/<p style=" margin-top:0px;
margin-bottom:0px;margin-left:0px; margin-right:0px; -qt-block-indent:0; text-indent:0px;
-qt-user-state:0;">PostgreSQL:Support 24x7 - Développement, Expertise et Formation