Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

From: pgsql-hackers-ow...@postgresql.org
> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Amit Kapila
> I think it beginning of segment (aka the first page of the segment), even
> the comment indicates the same.
> 
> /*
> * Whenever switching to a new WAL segment, we read the first page of
> * the file and validate its header, even if that's not where the
> * target record is. ...
> ..
> */
> 
> However, on again looking at the code, it seems that part of code behaves
> similarly for both 9.2 and 9.3.

Yes, the code behaves similarly in 9.2 and later.  FYI, ValidXLogHeader() is 
called at two sites.  The earlier one checks the first page of a segment when 
the real target page is different, and the latter one checks any page including 
the first page of a segment.


> ..Because node3 found a WAL
> !  * record fragment at the end of segment 10, it expects to find the !
> * remaining fragment at the beginning of WAL segment 11 streamed from !
> * node2. But there was a fragment of a different WAL record, because !  *
> node2 overwrote a different WAL record at the end of segment 10 across !
> * to 11.
> 
> How does node3 ensure that the fragment of WAL in segment 11 is different?
> Isn't it possible that when node2 overwrites the last record in WAL segment
> 10, it writes a record of slightly different contents but which is of the
> same size as an original record in WAL segment 10?

That case is detected by checking the CRC value in the XLOG record header.

Regards
Takayuki Tsunakawa


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Amit Kapila]

On Tue, Nov 15, 2016 at 7:51 AM, Tsunakawa, Takayuki
 wrote:
> From: pgsql-hackers-ow...@postgresql.org
>> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Amit Kapila
>> It looks like the code in 9.3 or later version uses the recptr as the target
>> segment location
>> (targetSegmentPtr) whereas 9.2 uses recptr as beginning of segment (readOff
>> = 0;).  If above understanding is right then it will set different values
>> for latestPagePtr in 9.2 and 9.3 onwards code.
>>
>
> In 9.2, the relevant variable is not recptr but recaddr.  recaddr in 9.2 and 
> recptr in later releases point to the beginning of a page just read, which is 
> not always the beginning of the segment (targetSegmentPtr).
>

I think it beginning of segment (aka the first page of the segment),
even the comment indicates the same.

/*
* Whenever switching to a new WAL segment, we read the first page of
* the file and validate its header, even if that's not where the
* target record is. ...
..
*/

However, on again looking at the code, it seems that part of code
behaves similarly for both 9.2 and 9.3.


..Because node3 found a WAL
!  * record fragment at the end of segment 10, it expects to find the
!  * remaining fragment at the beginning of WAL segment 11 streamed from
!  * node2. But there was a fragment of a different WAL record, because
!  * node2 overwrote a different WAL record at the end of segment 10 across
!  * to 11.

How does node3 ensure that the fragment of WAL in segment 11 is
different?  Isn't it possible that when node2 overwrites the last
record in WAL segment 10, it writes a record of slightly different
contents but which is of the same size as an original record in WAL
segment 10?


-- 
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

From: pgsql-hackers-ow...@postgresql.org
> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Robert Haas
> Let me try to be more clear.  I will not commit this patch if it is not
> properly commented.  I doubt that anyone else will, either.
> 
> The fact that those code changes already exist in 9.4+ is not a reason to
> back-port them to earlier releases without a proper explanation of why we
> are doing it.  Very possibly, we should also improve the comments in newer
> branches so that future authors don't reintroduce whatever bugs were fixed
> by these changes.  But whether we do that or not, I am not going to commit
> uncommented patches to complex code in order to fix obscure bugs in
> 3+-year-old branches.  I think that is a non-starter.
> 

OK, although I'm not perfectly sure what to add as a comment, I added an 
example scenario as a comment because I thought a concrete situation helps to 
understand the existing two paragraphs.  Is this good?

Regards
Takayuki Tsunakawa



cascading_standby_stuck_v2.patch
Description: cascading_standby_stuck_v2.patch

-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

From: pgsql-hackers-ow...@postgresql.org
> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Amit Kapila
> It looks like the code in 9.3 or later version uses the recptr as the target
> segment location
> (targetSegmentPtr) whereas 9.2 uses recptr as beginning of segment (readOff
> = 0;).  If above understanding is right then it will set different values
> for latestPagePtr in 9.2 and 9.3 onwards code.
> 

In 9.2, the relevant variable is not recptr but recaddr.  recaddr in 9.2 and 
recptr in later releases point to the beginning of a page just read, which is 
not always the beginning of the segment (targetSegmentPtr).

Regards
Takayuki Tsunakawa


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Robert Haas]

On Thu, Nov 10, 2016 at 12:13 AM, Tsunakawa, Takayuki
 wrote:
> From: pgsql-hackers-ow...@postgresql.org
>> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Robert Haas
>> OK.  I agree that's a problem.  However, your patch adds zero new comment
>> text while removing some existing comments, so I can't easily tell how it
>> solves that problem or whether it does so correctly.  Even if I were smart
>> enough to figure it out, I wouldn't want to rely on the next person also
>> being that smart.  This is obviously a subtle problem in tricky code, so
>> a clear explanation of the fix seems like a very good idea.
>
> The comment describes what the code is trying to achieve.  Actually, I just 
> imitated the code and comment of later major releases.  The only difference 
> between later releases and my patch (for 9.2) is whether the state is stored 
> in XLogReaderStruct or as global variables.  Below is the comment from 9.6, 
> where the second paragraph describes what the two nested if conditions mean.  
> The removed comment lines are what became irrelevant, which is also not 
> present in later major releases.

Let me try to be more clear.  I will not commit this patch if it is
not properly commented.  I doubt that anyone else will, either.

The fact that those code changes already exist in 9.4+ is not a reason
to back-port them to earlier releases without a proper explanation of
why we are doing it.  Very possibly, we should also improve the
comments in newer branches so that future authors don't reintroduce
whatever bugs were fixed by these changes.  But whether we do that or
not, I am not going to commit uncommented patches to complex code in
order to fix obscure bugs in 3+-year-old branches.  I think that is a
non-starter.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Amit Kapila]

On Thu, Nov 10, 2016 at 10:43 AM, Tsunakawa, Takayuki
 wrote:
> From: pgsql-hackers-ow...@postgresql.org
>> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Robert Haas
>> OK.  I agree that's a problem.  However, your patch adds zero new comment
>> text while removing some existing comments, so I can't easily tell how it
>> solves that problem or whether it does so correctly.  Even if I were smart
>> enough to figure it out, I wouldn't want to rely on the next person also
>> being that smart.  This is obviously a subtle problem in tricky code, so
>> a clear explanation of the fix seems like a very good idea.
>
> The comment describes what the code is trying to achieve.  Actually, I just 
> imitated the code and comment of later major releases.  The only difference 
> between later releases and my patch (for 9.2) is whether the state is stored 
> in XLogReaderStruct or as global variables.  Below is the comment from 9.6, 
> where the second paragraph describes what the two nested if conditions mean.  
> The removed comment lines are what became irrelevant, which is also not 
> present in later major releases.
>
> /*
>  * Since child timelines are always assigned a TLI greater than their
>  * immediate parent's TLI, we should never see TLI go backwards across
>  * successive pages of a consistent WAL sequence.
>  *
>  * Sometimes we re-read a segment that's already been (partially) 
> read. So
>  * we only verify TLIs for pages that are later than the last 
> remembered
>  * LSN.
>  */
>

I think the changes which you are referring has been done as part of
commit 7fcbf6a405ffc12a4546a25b98592ee6733783fc.  There is no mention
of such a bug fix in that commit; however, it is quite possible that
such a change has fixed the problem you have reported.  It is not
clear if we can directly copy that change and it seems to me the
change copied is also not complete.  It looks like the code in 9.3 or
later version uses the recptr as the target segment location
(targetSegmentPtr) whereas 9.2 uses recptr as beginning of segment
(readOff = 0;).  If above understanding is right then it will set
different values for latestPagePtr in 9.2 and 9.3 onwards code.


-- 
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

From: pgsql-hackers-ow...@postgresql.org
> [mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Robert Haas
> OK.  I agree that's a problem.  However, your patch adds zero new comment
> text while removing some existing comments, so I can't easily tell how it
> solves that problem or whether it does so correctly.  Even if I were smart
> enough to figure it out, I wouldn't want to rely on the next person also
> being that smart.  This is obviously a subtle problem in tricky code, so
> a clear explanation of the fix seems like a very good idea.

The comment describes what the code is trying to achieve.  Actually, I just 
imitated the code and comment of later major releases.  The only difference 
between later releases and my patch (for 9.2) is whether the state is stored in 
XLogReaderStruct or as global variables.  Below is the comment from 9.6, where 
the second paragraph describes what the two nested if conditions mean.  The 
removed comment lines are what became irrelevant, which is also not present in 
later major releases.

/*
 * Since child timelines are always assigned a TLI greater than their
 * immediate parent's TLI, we should never see TLI go backwards across
 * successive pages of a consistent WAL sequence.
 *
 * Sometimes we re-read a segment that's already been (partially) read. 
So
 * we only verify TLIs for pages that are later than the last remembered
 * LSN.
 */

Regards
Takayuki Tsunakawa




-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Robert Haas]

On Thu, Aug 25, 2016 at 10:33 PM, Tsunakawa, Takayuki
 wrote:
> The processing went as follows.
>
> 1. node1's timeline is 140.  It wrote a WAL record at the end of WAL segment 
> 117.  The WAL record didn't fit the last page, so it was split across 
> segments 117 and 118.
>
> 2. WAL segment 117 was archived.
>
> 3. node1 got down, and node2 was promoted.
>
> 4. As part of the recovery process, node2 retrieves WAL segment 117 from 
> archive.  It found a WAL record fragment at the end of the segment but could 
> not find the remaining fragment in segment 118, so node2 stops recovery there.
>
> LOG:  restored log file "008C028C0075" from archive
> LOG:  received promote request
> LOG:  redo done at 28C/75FFF738
>
> 5. node2 becomes the primary, and its timeline becomes 118.  node3 is 
> disconnected by node2 (but later reconnectes to node2).
>
> LOG:  terminating all walsender processes to force cascaded standby(s) to 
> update timeline and reconnect
>
> 6. node3 retrieves and applies WAL segment 117 from archive.
>
> LOG:  restored log file "008C028C0075" from archive
>
> 7. node3 found .history file for time line 141 and renews its timeline to 141.
>
> 8. Because node3 found a WAL record fragment at the end of segment 117, it 
> expects to find the remaining fragment at the beginning of WAL segment 118 
> streamed from node2.  But there was a fragment of a different WAL record, 
> because node2 overwrote a different WAL record at the end of segment 117 
> across to 118.
>
> LOG:  invalid contrecord length 5892 in log file 652, segment 118, offset 0
>
> 9. node3 then retrieves segment 117 from archive again to get the WAL record 
> at the end of segment 117.  However, as node3's timeline is already 141, it 
> complains about the older timeline when it sees the timeline 140 at the 
> beginning of segment 117.
>
> LOG:  out-of-sequence timeline ID 140 (after 141) in log file 652, segment 
> 117, offset 0

OK.  I agree that's a problem.  However, your patch adds zero new
comment text while removing some existing comments, so I can't easily
tell how it solves that problem or whether it does so correctly.  Even
if I were smart enough to figure it out, I wouldn't want to rely on
the next person also being that smart.  This is obviously a subtle
problem in tricky code, so a clear explanation of the fix seems like a
very good idea.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

From: pgsql-hackers-ow...@postgresql.org
[mailto:pgsql-hackers-ow...@postgresql.org] On Behalf Of Michael Paquier
> 9.3 has addressed that by allowing streaming standbys to perform timeline
> jumps via the replication protocol. Doesn't this problem enter in this area?

IIUC, that new feature enables timeline switch without disconnecting the 
standby and without WAL archive.  I think 9.2 can perform timeline switch with 
WAL archive.  In fact, the customer said they hit the problem only once in many 
occurrences of the same test.  The bug seems to emerge depending on the timing.

Regards
Takayuki Tsunakawa




-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Re: [HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Michael Paquier]

On Fri, Aug 26, 2016 at 11:33 AM, Tsunakawa, Takayuki
 wrote:
> Our customer hit a problem of cascading replication, and we found the cause.  
> They are using the latest PostgreSQL 9.2.18.  The bug seems to have been 
> fixed in 9.4 and higher during the big modification of xlog.c, but it's not 
> reflected in older releases.
>
> The attached patch is for 9.2.18.  This just borrows the idea from 9.4 and 
> higher.
>
> But we haven't been able to reproduce the problem.  Could you review the 
> patch and help to test it?  I would very much appreciate it if you could 
> figure out how to reproduce the problem easily.
> [...]
> LOG:  out-of-sequence timeline ID 140 (after 141) in log file 652, segment 
> 117, offset 0

9.3 has addressed that by allowing streaming standbys to perform
timeline jumps via the replication protocol. Doesn't this problem
enter in this area?
-- 
Michael


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

[HACKERS] [bug fix] Cascading standby cannot catch up and get stuck emitting the same message repeatedly

[Tsunakawa, Takayuki]

Hello,

Our customer hit a problem of cascading replication, and we found the cause.  
They are using the latest PostgreSQL 9.2.18.  The bug seems to have been fixed 
in 9.4 and higher during the big modification of xlog.c, but it's not reflected 
in older releases.

The attached patch is for 9.2.18.  This just borrows the idea from 9.4 and 
higher.

But we haven't been able to reproduce the problem.  Could you review the patch 
and help to test it?  I would very much appreciate it if you could figure out 
how to reproduce the problem easily.


PROBLEM


The customer's configuration consists of three nodes: node1 is a primary, node2 
is a synchronous standby, and node3 is a cascading standby.  The primary 
archives WAL to a shared (network?) storage and the standbys read archived WAL 
from there with restore_command.  recovery_target_timeline is set to 'latest' 
on the standbys.

When node1 dies and node2 is promoted to a primary, node3 cannot catch up node2 
forever, emitting the following message repeatedly:

LOG:  out-of-sequence timeline ID 140 (after 141) in log file 652, segment 117, 
offset 0

The expected behavior is that node3 catches up node2 and keeps synchronization.


CAUSE


The processing went as follows.

1. node1's timeline is 140.  It wrote a WAL record at the end of WAL segment 
117.  The WAL record didn't fit the last page, so it was split across segments 
117 and 118.

2. WAL segment 117 was archived.

3. node1 got down, and node2 was promoted.

4. As part of the recovery process, node2 retrieves WAL segment 117 from 
archive.  It found a WAL record fragment at the end of the segment but could 
not find the remaining fragment in segment 118, so node2 stops recovery there.

LOG:  restored log file "008C028C0075" from archive
LOG:  received promote request
LOG:  redo done at 28C/75FFF738

5. node2 becomes the primary, and its timeline becomes 118.  node3 is 
disconnected by node2 (but later reconnectes to node2).

LOG:  terminating all walsender processes to force cascaded standby(s) to 
update timeline and reconnect

6. node3 retrieves and applies WAL segment 117 from archive.

LOG:  restored log file "008C028C0075" from archive

7. node3 found .history file for time line 141 and renews its timeline to 141.

8. Because node3 found a WAL record fragment at the end of segment 117, it 
expects to find the remaining fragment at the beginning of WAL segment 118 
streamed from node2.  But there was a fragment of a different WAL record, 
because node2 overwrote a different WAL record at the end of segment 117 across 
to 118.

LOG:  invalid contrecord length 5892 in log file 652, segment 118, offset 0

9. node3 then retrieves segment 117 from archive again to get the WAL record at 
the end of segment 117.  However, as node3's timeline is already 141, it 
complains about the older timeline when it sees the timeline 140 at the 
beginning of segment 117.

LOG:  out-of-sequence timeline ID 140 (after 141) in log file 652, segment 117, 
offset 0



Regards
Takayuki Tsunakawa



cascading_standby_stuck.patch
Description: cascading_standby_stuck.patch

-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers