Обсуждение: Longer and longer updates
Hi all-
In doing some stress testing I noticed that our database was slowing
down significantly after doing the same update repeatedly on a table
with no indexes, so I tried this:
1) created an empty database
2) made a simple table with a single row:
create table test (key int4, data int4);
insert into test values (1, 0);
3) executed this statement tons of times:
update test set data=1234 where key=1
Here are the results -- it's pretty discouraging, I hope I'm making some
simple mistake, or maybe this is expected behavior for some reason?
After this many updates ...it took this long for 1000 more updates
----------------------- ------------------------------------------
0 10880 ms
5,000 10549 ms
10,000 17380 ms
15,000 20040 ms
20,000 20060 ms
25,000 20589 ms
30,000 30749 ms
35,000 30350 ms
40,000 30910 ms
45,000 37570 ms
50,000 40379 ms
This seems to be independent of starting and stopping my client and the
postmaster, running vacuum, praying, etc. I'm on RedHat6.2
running with the 7.1beta4 rpms.
Anyone know what's going on here?
Steve Wilmarth
* Steve Wilmarth <swilmarth@eknow.com> [010205 12:50] wrote: > > Hi all- > > In doing some stress testing I noticed that our database was slowing > down significantly after doing the same update repeatedly on a table > with no indexes, so I tried this: > > 1) created an empty database > > 2) made a simple table with a single row: > > create table test (key int4, data int4); > insert into test values (1, 0); > > 3) executed this statement tons of times: > > update test set data=1234 where key=1 > > Here are the results -- it's pretty discouraging, I hope I'm making some > simple mistake, or maybe this is expected behavior for some reason? > > After this many updates ...it took this long for 1000 more updates > ----------------------- ------------------------------------------ > 0 10880 ms > 5,000 10549 ms > 10,000 17380 ms > 15,000 20040 ms > 20,000 20060 ms > 25,000 20589 ms > 30,000 30749 ms > 35,000 30350 ms > 40,000 30910 ms > 45,000 37570 ms > 50,000 40379 ms > > This seems to be independent of starting and stopping my client and the > postmaster, running vacuum, praying, etc. I'm on RedHat6.2 > running with the 7.1beta4 rpms. > > Anyone know what's going on here? You're not vacuuming. -- -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] "I have the heart of a child; I keep it in a jar on my desk."
if you're doing updates in a single transaction, you'll realize speed gains by distributing the updates into multiple transactions. postgres won't have to keep multiple copies that way. hth kapil > > > > 3) executed this statement tons of times: > > > > update test set data=1234 where key=1 > > > > Here are the results -- it's pretty discouraging, I hope I'm making some > > simple mistake, or maybe this is expected behavior for some reason? > > > > After this many updates ...it took this long for 1000 more updates > > ----------------------- ------------------------------------------ > > 0 10880 ms > > 5,000 10549 ms > > 10,000 17380 ms > > 15,000 20040 ms > > 20,000 20060 ms > > 25,000 20589 ms > > 30,000 30749 ms > > 35,000 30350 ms > > 40,000 30910 ms > > 45,000 37570 ms > > 50,000 40379 ms > > > > This seems to be independent of starting and stopping my client and the > > postmaster, running vacuum, praying, etc. I'm on RedHat6.2 > > running with the 7.1beta4 rpms.
On Mon, 5 Feb 2001, Alfred Perlstein wrote: > > ... > > 3) executed this statement tons of times: > > > > update test set data=1234 where key=1 > > > > Here are the results -- it's pretty discouraging, I hope I'm making some > > simple mistake, or maybe this is expected behavior for some reason? > > > > After this many updates ...it took this long for 1000 more updates > > ----------------------- ------------------------------------------ > > 0 10880 ms > > 5,000 10549 ms > > 10,000 17380 ms > > 15,000 20040 ms > > 20,000 20060 ms > > 25,000 20589 ms > > 30,000 30749 ms > > 35,000 30350 ms > > 40,000 30910 ms > > 45,000 37570 ms > > 50,000 40379 ms > > > > This seems to be independent of starting and stopping my client and the > > postmaster, running vacuum, praying, etc. I'm on RedHat6.2 > > running with the 7.1beta4 rpms. > > > > Anyone know what's going on here? > > You're not vacuuming. VACUUMing shows some improvement -- it takes me back to the 20000ms level or so, but I can never seem to get back to the 10000ms. Is this what you'd expect? Also, any thoughts on how often to run VACUUM? The manual recommends "nightly" for an operational database, but mine was slowing down significantly after only a couple of hours of use. Are there any rules of thumb people use? Also, VACUUM can take a few seconds to run -- is that something people just live with and let the waiting transactions pile up? Or do you do something like VACUUM a single table at a time and rotate which table gets done? Is that a reasonable thing to do? Thanks very much for the help- Steve Wilmarth