Re: Online base backup from the hot-standby - Mailing list pgsql-hackers

On 12-01-17 05:38 AM, Fujii Masao wrote: <blockquote
cite="mid:CAHGQGwEXK39yByx6Q25u6r-Z4GZg0VEvtw6tvUpLBBvchZEUEw@mail.gmail.com"type="cite"><pre wrap="">On Fri, Jan 13,
2012at 5:02 PM, Fujii Masao <a class="moz-txt-link-rfc2396E"
href="mailto:masao.fujii@gmail.com"><masao.fujii@gmail.com></a>wrote:
 
</pre><blockquote type="cite"><pre wrap="">The amount of code changes to allow pg_basebackup to make a backup from
the standby seems to be small. So I ended up merging that changes and the
infrastructure patch. WIP patch attached. But I'd happy to split the patch again
if you want.
</pre></blockquote><pre wrap="">
Attached is the updated version of the patch. I wrote the limitations of
standby-only backup in the document and changed the error messages.

</pre></blockquote><br /> Here is my review of this verison of the patch. I think this patch has been in every CF for
9.2and I feel it is getting close to being committed.  The only issue of significants is a crash I encountered while
testing,see below.   <br /><br /> I am fine with including the pg_basebackup changes in the patch it also makes testing
someof the other changes possible.<br /><br /><br /> The documentation updates you have are good<br /><br /> I don't
seeany issues looking at the code.<br /><br /><br /><br /> Testing Review<br /> --------------------------------<br
/><br/> I encountered this on my first replica (the one based on the master).  I am not sure if it is related to this
patch,it happened after the pg_basebackup against the replica finished.<br /><br /> TRAP: FailedAssertion("!(((xid) !=
((TransactionId)0)))", File: "twophase.c", Line: 1238)<br /> LOG:  startup process (PID 12222) was terminated by signal
6:Aborted<br /> LOG:  terminating any other active server processes<br /><br /> A little earlier this postmaster had
printed.<br/><br /> LOG:  restored log file "00000001000000000000001F" from archive<br /> LOG:  restored log file
"000000010000000000000020"from archive<br /> cp: cannot stat `/usr/local/pgsql92git/archive/000000010000000000000021':
Nosuch file or directory<br /> LOG:  unexpected pageaddr 0/19000000 in log file 0, segment 33, offset 0<br /> cp:
cannotstat `/usr/local/pgsql92git/archive/000000010000000000000021': No such file or directory<br /><br /><br /> I have
NOTbeen able to replicate this error  and I am not sure exactly what I had done in my testing prior to that point.  <br
/><br/><br /> In another test run I had<br /><br /> - set full page writes=off and did a checkpoint<br /> - Started the
pg_basebackup<br/> - set full_page_writes=on and did a HUP + some database activity that might have forced a
checkpoint.<br/><br /> I got this message from pg_basebackup.<br /> ./pg_basebackup -D ../data3 -l foo -h localhost -p
5438<br /> pg_basebackup: could not get WAL end position from server<br /><br /> I point this out because the message
isdifferent than the normal "could not initiate base backup: FATAL:  WAL generated with full_page_writes=off" thatI
normallysee.    We might want to add a PQerrorMessage(conn)) to pg_basebackup to print the error details.  Since this
patchdidn't actually change pg_basebackup I don't think your required to improve the error messages in it.  I am just
mentioningthis because it came up in testing.<br /><br /><br /> The rest of the tests I did involving changing
full_page_writes with/without checkpoints and sighups and promoting the replica seemed to work as expected.<br /><br
/><br/><br /><br /><blockquote cite="mid:CAHGQGwEXK39yByx6Q25u6r-Z4GZg0VEvtw6tvUpLBBvchZEUEw@mail.gmail.com"
type="cite"><prewrap="">Regards,
 

</pre> <pre wrap="">
<fieldset class="mimeAttachmentHeader"></fieldset>

</pre></blockquote><br />