On Fri, Dec 9, 2016 at 4:54 AM, Robert Haas wrote:
> On Wed, Dec 7, 2016 at 11:20 PM, Michael Paquier
> wrote:
>> OK, I rewrote a bit the patch as attached. What do you think?
>
> Committed and back-patched all the way back to 9.2.
Thanks!
>>> Right (I think). If we set and clear delayChkpt ar
On Wed, Dec 7, 2016 at 11:20 PM, Michael Paquier
wrote:
> OK, I rewrote a bit the patch as attached. What do you think?
Committed and back-patched all the way back to 9.2.
>> Right (I think). If we set and clear delayChkpt around this work, we
>> don't need the immediate sync.
>
> My point is a
On Thu, Dec 8, 2016 at 6:03 AM, Robert Haas wrote:
> Michael, your greetings were passed on to me with a request that I
> look at this thread.
Thanks for showing up!
> On Fri, Nov 18, 2016 at 12:33 PM, Michael Paquier
> wrote:
More seriously, if there could be more details regarding that,
Michael, your greetings were passed on to me with a request that I
look at this thread.
On Fri, Nov 18, 2016 at 12:33 PM, Michael Paquier
wrote:
>>> More seriously, if there could be more details regarding that, I would
>>> think that we could say something like "logging the init fork is
>>> mand
On Thu, Nov 17, 2016 at 7:06 AM, Robert Haas wrote:
> Yeah, but surely it's obvious that if you don't WAL log it, it's not
> going to work for archiving or streaming. It's a lot less obvious why
> you have to WAL log it when you're not doing either of those things if
> you've already written it a
On Wed, Nov 16, 2016 at 11:55 PM, Michael Paquier
wrote:
> On Wed, Nov 16, 2016 at 7:09 PM, Robert Haas wrote:
>> On Wed, Nov 16, 2016 at 3:54 PM, Michael Paquier
>> wrote:
>>> Indeed I missed this comment block. Please let me suggest the following
>>> instead:
>>> /*
>>> * Set up an init fo
On Wed, Nov 16, 2016 at 7:09 PM, Robert Haas wrote:
> On Wed, Nov 16, 2016 at 3:54 PM, Michael Paquier
> wrote:
>> Indeed I missed this comment block. Please let me suggest the following
>> instead:
>> /*
>> * Set up an init fork for an unlogged table so that it can be correctly
>> - * reinit
On Wed, Nov 16, 2016 at 3:54 PM, Michael Paquier
wrote:
> On Wed, Nov 16, 2016 at 11:45 AM, Robert Haas wrote:
>> The header comment for heap_create_init_fork() says this:
>>
>> /*
>> * Set up an init fork for an unlogged table so that it can be correctly
>> * reinitialized on restart. Since w
On Wed, Nov 16, 2016 at 11:45 AM, Robert Haas wrote:
> The header comment for heap_create_init_fork() says this:
>
> /*
> * Set up an init fork for an unlogged table so that it can be correctly
> * reinitialized on restart. Since we're going to do an immediate sync, we
> * only need to xlog th
On Thu, Nov 10, 2016 at 9:25 PM, Michael Paquier
wrote:
> On Thu, Nov 10, 2016 at 10:52 PM, Kuntal Ghosh
> wrote:
>> On Thu, Nov 10, 2016 at 3:42 PM, Michael Paquier
>> wrote:
>>> Nah. Looking at the code the fix is quite obvious.
>>> heap_create_init_fork() is checking for XLogIsNeeded() to dec
On Thu, Nov 10, 2016 at 10:52 PM, Kuntal Ghosh
wrote:
> On Thu, Nov 10, 2016 at 3:42 PM, Michael Paquier
> wrote:
>> Nah. Looking at the code the fix is quite obvious.
>> heap_create_init_fork() is checking for XLogIsNeeded() to decide if
>> the INIT forknum should be logged or not. But this is w
On Thu, Nov 10, 2016 at 3:42 PM, Michael Paquier
wrote:
> On Thu, Nov 10, 2016 at 5:15 PM, Michael Paquier
> wrote:
>> Okay, so what happens is that the CREATE TABLESPACE record removes the
>> tablespace directory and recreates a fresh one, but as no CREATE
>> records are created for unlogged tab
On Thu, Nov 10, 2016 at 5:15 PM, Michael Paquier
wrote:
> Okay, so what happens is that the CREATE TABLESPACE record removes the
> tablespace directory and recreates a fresh one, but as no CREATE
> records are created for unlogged tables the init fork is not
> re-created. It seems to me that we sh
On Thu, Nov 10, 2016 at 4:33 PM, Michael Paquier
wrote:
> On Thu, Nov 10, 2016 at 4:23 PM, konstantin knizhnik
> wrote:
>> No, it is latest sources from Postgres repository.
>> Please notice that you should create new database and tablespace to
>> reproduce this issue.
>> So actually the whole s
On Thu, Nov 10, 2016 at 4:23 PM, konstantin knizhnik
wrote:
> No, it is latest sources from Postgres repository.
> Please notice that you should create new database and tablespace to reproduce
> this issue.
> So actually the whole sequence is
>
> mkdir fs
> initdb -D pgsql
> pg_ctl -D pgsql -l lo
On Nov 10, 2016, at 10:17 AM, Michael Paquier wrote:
>
> Hm.. I cannot reproduce what you see on Linux or macos. Perhaps you
> have locally a standby pointing as well to this tablespace?
No, it is latest sources from Postgres repository.
Please notice that you should create new database and tab
On Thu, Nov 10, 2016 at 3:29 AM, Robert Haas wrote:
> On Wed, Nov 9, 2016 at 11:56 AM, Konstantin Knizhnik
> wrote:
>> Now simulate server crash using using "pkill -9 postgres".
>>
>> knizhnik@knizhnik:~/dtm-data$ rm -f logfile ; pg_ctl -D pgsql.master -l
>> logfile start
>> pg_ctl: another serve
On Wed, Nov 9, 2016 at 11:56 AM, Konstantin Knizhnik
wrote:
> Now simulate server crash using using "pkill -9 postgres".
>
> knizhnik@knizhnik:~/dtm-data$ rm -f logfile ; pg_ctl -D pgsql.master -l
> logfile start
> pg_ctl: another server might be running; trying to start server anyway
> server sta
Andres Freund wrote:
> On 2014-11-19 11:26:56 +, Albe Laurenz wrote:
>> I observed an interesting (and I think buggy) behaviour today after one of
>> our clusters crashed due to an "out of space" condition in the data
>> directory.
>
> Hah, just a couple days I pushed a fix for that ;)
>
> h
Hi,
On 2014-11-19 11:26:56 +, Albe Laurenz wrote:
> I observed an interesting (and I think buggy) behaviour today after one of
> our clusters crashed due to an "out of space" condition in the data directory.
Hah, just a couple days I pushed a fix for that ;)
http://archives.postgresql.org/me
Hi Heikki,
On Mon, Feb 11, 2013 at 7:28 PM, Heikki Linnakangas wrote:
> On 11.02.2013 08:44, Jeevan Chalke wrote:
>
>> Hi,
>>
>> Any review comments on this ?
>>
>
> Sorry for the delay.
>
> I did some minor cleanup on this. I added code to pg_resetxlog and
> pg_controldata to reset / display t
On 11.02.2013 08:44, Jeevan Chalke wrote:
Hi,
Any review comments on this ?
Sorry for the delay.
I did some minor cleanup on this. I added code to pg_resetxlog and
pg_controldata to reset / display the current unlogged LSN value. I
moved the static counter, for temporary relations, back to
Hi,
Any review comments on this ?
On Tue, Jan 29, 2013 at 6:03 PM, Jeevan Chalke <
jeevan.cha...@enterprisedb.com> wrote:
> Hi Heikki,
>
>
> On Mon, Jan 28, 2013 at 2:34 PM, Heikki Linnakangas <
> hlinnakan...@vmware.com> wrote:
>
>> On 23.01.2013 17:30, Robert Haas wrote:
>>
>>> On Wed, Jan 23
Hi Heikki,
On Mon, Jan 28, 2013 at 2:34 PM, Heikki Linnakangas wrote:
> On 23.01.2013 17:30, Robert Haas wrote:
>
>> On Wed, Jan 23, 2013 at 4:04 AM, Jeevan Chalke
>> >
>> wrote:
>>
>>> I guess my earlier patch, which was directly incrementing
>>>
>>> ControlFile->unloggedLSN counter was the c
On Mon, Jan 28, 2013 at 4:04 AM, Heikki Linnakangas
wrote:
> Do we need to do anything to unloggedLSN in pg_resetxlog?
Does the server go into recovery after pg_resetxlog? If so, no. If
not, probably, but I have no idea what. There's no "safe" value in
that case; what we ought to do is force a
On 23.01.2013 17:30, Robert Haas wrote:
On Wed, Jan 23, 2013 at 4:04 AM, Jeevan Chalke
wrote:
I guess my earlier patch, which was directly incrementing
ControlFile->unloggedLSN counter was the concern as it will take
ControlFileLock several times.
In this version of patch I did what Robert ha
On Wed, Jan 23, 2013 at 4:04 AM, Jeevan Chalke
wrote:
> Yes.
>
> I guess my earlier patch, which was directly incrementing
> ControlFile->unloggedLSN counter was the concern as it will take
> ControlFileLock several times.
>
> In this version of patch I did what Robert has suggested. At start of t
On Wed, Jan 16, 2013 at 3:24 AM, Robert Haas wrote:
> On Tue, Jan 15, 2013 at 4:26 PM, Tom Lane wrote:
> > Robert Haas writes:
> >> I think that might be acceptable from a performance point of view -
> >> after all, if the index is unlogged, you're saving the cost of WAL -
> >> but I guess I st
On Tue, Jan 15, 2013 at 4:26 PM, Tom Lane wrote:
> Robert Haas writes:
>> I think that might be acceptable from a performance point of view -
>> after all, if the index is unlogged, you're saving the cost of WAL -
>> but I guess I still prefer a generic solution to this problem (a
>> generalizati
Robert Haas writes:
> I think that might be acceptable from a performance point of view -
> after all, if the index is unlogged, you're saving the cost of WAL -
> but I guess I still prefer a generic solution to this problem (a
> generalization of GetXLogRecPtrForTemp) rather than a special-purpos
On Tue, Jan 15, 2013 at 1:58 PM, Heikki Linnakangas
wrote:
>> I think what Heikki had in mind was that the copy in the index would be
>> the authoritative one, not some image in shared memory. This'd imply
>> dirtying the root page on every insert, as well as increased contention
>> for the root
On 2013-01-15 20:58:00 +0200, Heikki Linnakangas wrote:
> On 15.01.2013 20:48, Tom Lane wrote:
> >Robert Haas writes:
> >>On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
> >> wrote:
> >>>Could we stash the counter e.g. in the root page of the index?
> >
> >>That would require maintaining a co
On 15.01.2013 20:48, Tom Lane wrote:
Robert Haas writes:
On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
wrote:
Could we stash the counter e.g. in the root page of the index?
That would require maintaining a counter per table rather than a
single global counter, which would be bad bec
On Tue, Jan 15, 2013 at 1:48 PM, Tom Lane wrote:
> Robert Haas writes:
>> On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
>> wrote:
>>> Could we stash the counter e.g. in the root page of the index?
>
>> That would require maintaining a counter per table rather than a
>> single global counte
Robert Haas writes:
> On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
> wrote:
>> Could we stash the counter e.g. in the root page of the index?
> That would require maintaining a counter per table rather than a
> single global counter, which would be bad because then we'd need to
> store on
On 15.01.2013 20:33, Robert Haas wrote:
On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
Could we stash the counter e.g. in the root page of the index?
That would require maintaining a counter per table rather than a
single global counter, which would be bad because then we'd need to
store
On Tue, Jan 15, 2013 at 1:10 PM, Heikki Linnakangas
wrote:
> On 15.01.2013 08:54, Jeevan Chalke wrote:
>>
>> For (2), I have added a new function called, GetXLogRecPtrForUnloogedRel()
>> which returns a fake LSN for GiST indexes. However, I have removed
>> GetXLogRecPtrForTemp() function and added
On 15.01.2013 08:54, Jeevan Chalke wrote:
For (2), I have added a new function called, GetXLogRecPtrForUnloogedRel()
which returns a fake LSN for GiST indexes. However, I have removed
GetXLogRecPtrForTemp() function and added its functionality inside this new
function itself to avoid complexity.
Hi Robert / Tom
I think to support GiST with unlogged table we need to do three things:
1. Teach the buffer manager that the LSN of a page not marked
BM_PERMANENT can be ignored
2. Teach GetXLogRecPtrForTemp() to allocate fake LSNs for GiST buffers
using a 64-bit, counter that is persisted acros
On Fri, Jan 6, 2012 at 7:52 AM, Heikki Linnakangas
wrote:
> In BufferSync(), we have this:
>
>> /*
>> * Unless this is a shutdown checkpoint, we write only permanent,
>> dirty
>> * buffers. But at shutdown time, we write all dirty buffers.
>> */
>> if (!(flag
On Tue, May 3, 2011 at 4:54 PM, Tom Lane wrote:
> Simon Riggs writes:
>> You're assuming that there are referential links *from* other tables
>> to the table with damage. In which case you would be correct. But of
>> course, if you needed that data for integrity you would never do that,
>> so the
Simon Riggs writes:
> You're assuming that there are referential links *from* other tables
> to the table with damage. In which case you would be correct. But of
> course, if you needed that data for integrity you would never do that,
> so the problem is a nonexistent use case. The suggested mode
On Tue, May 3, 2011 at 8:21 PM, Simon Riggs wrote:
> The current assessment is that UNLOGGED tables are useful only for
> running a data cache. If the database crashes, then the table is
> truncated and you must refill the cache. If that is the case, then it
> must surely be better to have a cache
On Tue, Apr 26, 2011 at 8:49 AM, Leonardo Francalanci wrote:
>> > If that 1% is random (not time/transaction related), usually you'd rather
>>have an empty table.
>>
>> Why do you think it would be random?
>
> "Heap blocks would be zeroed if they were found to be damaged, following a
> crash."
>
2011/4/25 Christopher Browne :
> On Mon, Apr 25, 2011 at 2:03 PM, Jesper Krogh wrote:
>> On 2011-04-25 20:00, Leonardo Francalanci wrote:
The amount of data loss on a big table will be <1% of the data
loss caused by truncating the whole table.
>>>
>>> If that 1% is random (not time/trans
> > If that 1% is random (not time/transaction related), usually you'd rather
>have an empty table.
>
> Why do you think it would be random?
"Heap blocks would be zeroed if they were found to be damaged, following a
crash."
If you erase full blocks, you have no idea what data you erased; it
Robert Haas writes:
> However, I think that the "logged table, unlogged index" idea is
> probably the most promising thing to think about doing first.
+1 for that --- it's clean, has a clear use-case, and would allow us
to manage the current mess around hash indexes more cleanly.
That is, hash in
On Mon, Apr 25, 2011 at 2:21 PM, Simon Riggs wrote:
>> Right, but the trick is how you identify which blocks you need to
>> zero. You used the word "damaged", which to me implied that the block
>> had been modified in some way but ended up with other than the
>> expected contents, so that somethi
On Mon, Apr 25, 2011 at 1:42 PM, Robert Haas wrote:
> On Mon, Apr 25, 2011 at 5:04 AM, Simon Riggs wrote:
>> On Mon, Apr 25, 2011 at 8:14 AM, Robert Haas wrote:
>>> On Apr 24, 2011, at 1:22 PM, Simon Riggs wrote:
Unlogged tables are a good new feature.
>>>
>>> Thanks.
>>>
I noticed Br
On Mon, Apr 25, 2011 at 7:00 PM, Leonardo Francalanci wrote:
>> The amount of data loss on a big
>> table will be <1% of the data loss
>>caused by truncating the whole table.
>
> If that 1% is random (not time/transaction related), usually you'd rather
> have an empty table.
Why do you think it
On Mon, Apr 25, 2011 at 2:03 PM, Jesper Krogh wrote:
> On 2011-04-25 20:00, Leonardo Francalanci wrote:
>>> The amount of data loss on a big table will be <1% of the data
>>> loss caused by truncating the whole table.
>>
>> If that 1% is random (not time/transaction related), usually you'd
>> rath
Jesper Krogh wrote:
> On 2011-04-25 20:00, Leonardo Francalanci wrote:
>>> The amount of data loss on a big table will be <1% of the data
>>> loss caused by truncating the whole table.
>>
>> If that 1% is random (not time/transaction related), usually
>> you'd rather have an empty table. In othe
On 2011-04-25 20:00, Leonardo Francalanci wrote:
> The amount of data loss on a big table will be <1% of the data
> loss caused by truncating the whole table.
If that 1% is random (not time/transaction related), usually you'd
rather have an empty table. In other words: is a table that is not
> The amount of data loss on a big
> table will be <1% of the data loss
>caused by truncating the whole table.
If that 1% is random (not time/transaction related), usually you'd rather have
an empty table. In other words: is a table that is not consistant with anything
else in the db useful?
--
On Mon, Apr 25, 2011 at 3:36 AM, Leonardo Francalanci wrote:
> The only data we can't rebuild it's the heap. So what about an option for
> UNlogged indexes on a LOGged table? It would always preserve data, and it
> would 'only' cost a rebuilding of the indexes in case of an unclean shutdown.
>
On Mon, Apr 25, 2011 at 5:04 AM, Simon Riggs wrote:
> On Mon, Apr 25, 2011 at 8:14 AM, Robert Haas wrote:
>> On Apr 24, 2011, at 1:22 PM, Simon Riggs wrote:
>>> Unlogged tables are a good new feature.
>>
>> Thanks.
>>
>>> I noticed Bruce had mentioned they were the equivalent of NoSQL, which
>>>
On Mon, Apr 25, 2011 at 8:14 AM, Robert Haas wrote:
> On Apr 24, 2011, at 1:22 PM, Simon Riggs wrote:
>> Unlogged tables are a good new feature.
>
> Thanks.
>
>> I noticed Bruce had mentioned they were the equivalent of NoSQL, which
>> I don't really accept.
>
> Me neither. I thought that was poo
On Mon, Apr 25, 2011 at 8:36 AM, Leonardo Francalanci wrote:
> The only data we can't rebuild it's the heap. So what about an option for
> UNlogged indexes on a LOGged table? It would always preserve data, and it
> would 'only' cost a rebuilding of the indexes in case of an unclean shutdown.
>
The only data we can't rebuild it's the heap. So what about an option for
UNlogged indexes on a LOGged table? It would always preserve data, and it would
'only' cost a rebuilding of the indexes in case of an unclean shutdown. I think
it would give a boost in performance for all those cases where
On Apr 24, 2011, at 1:22 PM, Simon Riggs wrote:
> Unlogged tables are a good new feature.
Thanks.
> I noticed Bruce had mentioned they were the equivalent of NoSQL, which
> I don't really accept.
Me neither. I thought that was poorly said.
> Heap blocks would be zeroed if they were found to be
On Sun, Apr 24, 2011 at 11:15 PM, Simon Riggs wrote:
> IMHO deleting all the data is a surprising option that will
> cause many people to curse us. I don't see preserving some of the data
> as being worse.
What possible damage to you want to recover from?
Without WAL logging after a software cra
On Sun, Apr 24, 2011 at 7:41 PM, Greg Stark wrote:
> On Sun, Apr 24, 2011 at 6:22 PM, Simon Riggs wrote:
>> My implementation path for that would be to add a crash_number onto
>> pg_control and pg_index. Any index marked as "unlogged, persistent"
>> would only be usable if it's crash number is th
On Sun, Apr 24, 2011 at 6:15 PM, Simon Riggs wrote:
> For large sets of low value data, it makes sense. Deleting all data,
> just simply because some of it might be damaged, is not the only
> option. IMHO deleting all the data is a surprising option that will
> cause many people to curse us. I don
On Sun, Apr 24, 2011 at 10:27 PM, Tom Lane wrote:
> Simon Riggs writes:
>> If you don't log changes to tables you have two choices if we crash
>> 1) truncate the table and any indexes
>> 2) rebuild any indexes damaged by the crash
>
> No, you have only one choice, and that's (1), because there's
Simon Riggs writes:
> If you don't log changes to tables you have two choices if we crash
> 1) truncate the table and any indexes
> 2) rebuild any indexes damaged by the crash
No, you have only one choice, and that's (1), because there's no
guarantee that what's in the table file is meaningful.
On Sun, Apr 24, 2011 at 6:22 PM, Simon Riggs wrote:
> My implementation path for that would be to add a crash_number onto
> pg_control and pg_index. Any index marked as "unlogged, persistent"
> would only be usable if it's crash number is the same as current
> system crash number.
>
> REINDEX woul
On Tue, Feb 22, 2011 at 8:03 AM, Cédric Villemain
wrote:
> The fact that you didn''t saw it might be enough to reconsider the way
> we display the unlogged state (and temp state) of a relation.
>
> Maybe some a "Durability: normal, temp, unlogged" line at bottom of
> the \d output ?
The term we
2011/2/22 Itagaki Takahiro :
> On Tue, Feb 22, 2011 at 18:11, Cédric Villemain
> wrote:
>> 2011/2/22 Itagaki Takahiro :
>>> psql \d(+) doesn't show any information about UNLOGGED and TEMP attributes
>>> for the table. So, we cannot know the table is unlogged or not unless
>>> we directly select fr
On Tue, Feb 22, 2011 at 18:11, Cédric Villemain
wrote:
> 2011/2/22 Itagaki Takahiro :
>> psql \d(+) doesn't show any information about UNLOGGED and TEMP attributes
>> for the table. So, we cannot know the table is unlogged or not unless
>> we directly select from pg_class.relpersistence. Is this
2011/2/22 Itagaki Takahiro :
> psql \d(+) doesn't show any information about UNLOGGED and TEMP attributes
> for the table. So, we cannot know the table is unlogged or not unless
> we directly select from pg_class.relpersistence. Is this a TODO item?
>
> The same issue is in TEMP tables, but we can
On Mon, Dec 27, 2010 at 3:22 AM, Simon Riggs wrote:
> Not sure from reading the docs whether unlogged indexes are supported on
> logged tables? Could you clarify (or clarify more often)? Does this
> solve the hash index situation?
They are not. The only place you'll see that the grammar has been
On Fri, 2010-12-24 at 23:36 -0500, Robert Haas wrote:
> Here's an updated patch for unlogged tables, incorporating the
> following changes since v4:
Looks good
> 5. Support unlogged GIN indexes.
Not sure from reading the docs whether unlogged indexes are supported on
logged tables? Could you cla
Alvaro Herrera writes:
> Excerpts from Robert Haas's message of sáb dic 18 02:21:41 -0300 2010:
>> 1. pg_dump needs an option to control whether unlogged tables are
>> dumped. --no-unlogged-tables seems like the obvious choice, assuming
>> we want the default to be to dump them, which seems like
On Mon, Dec 20, 2010 at 9:05 AM, Alvaro Herrera
wrote:
> Excerpts from Robert Haas's message of sáb dic 18 02:21:41 -0300 2010:
>> Here's an attempt to summarize the remaining issues with this patch
>> that I know about. I may have forgotten something, so please mention
>> it if you notice someth
Excerpts from Robert Haas's message of sáb dic 18 02:21:41 -0300 2010:
> Here's an attempt to summarize the remaining issues with this patch
> that I know about. I may have forgotten something, so please mention
> it if you notice something missing.
>
> 1. pg_dump needs an option to control wheth
On Sat, Dec 18, 2010 at 12:27 PM, Kevin Grittner
wrote:
> Robert Haas wrote:
>
>> If there's any third-party code out there that is checking
>> rd_istemp, it likely also needs to be revised to check whether
>> WAL-logging is needed, not whether the relation is temp. The way
>> I've coded it, such
Robert Haas wrote:
> If there's any third-party code out there that is checking
> rd_istemp, it likely also needs to be revised to check whether
> WAL-logging is needed, not whether the relation is temp. The way
> I've coded it, such code will fail to compile, and can be very
> easily fixed by s
Here's an attempt to summarize the remaining issues with this patch
that I know about. I may have forgotten something, so please mention
it if you notice something missing.
1. pg_dump needs an option to control whether unlogged tables are
dumped. --no-unlogged-tables seems like the obvious choic
On Fri, Dec 17, 2010 at 4:17 PM, Tom Lane wrote:
>> Since these bits will only be set/cleared when the buffer mapping is
>> changed, can we examine this bit without taking the spinlock?
>
> Only if you're willing for the result to be unreliable.
If we read the bits while someone else is writing t
[ hit send too soon ... ]
Robert Haas writes:
> Since these bits will only be set/cleared when the buffer mapping is
> changed, can we examine this bit without taking the spinlock?
Only if you're willing for the result to be unreliable. In the
case of the xlog flush bit, I don't believe an extr
Robert Haas writes:
> Allright, what do you want to call the other bit, then? BM_SKIP_XLOG_FLUSH?
Like I said, I'd be tempted to invert the sense, so that the flag is set
for normal relations. Then it becomes something like BM_FLUSH_XLOG.
regards, tom lane
--
Sent via
On Fri, Dec 17, 2010 at 3:15 PM, Tom Lane wrote:
> Robert Haas writes:
>> On Fri, Dec 17, 2010 at 3:03 PM, Tom Lane wrote:
>>> Yeah. I think that BM_UNLOGGED might be a poor choice for the flag name,
>>> just because it overstates what the bufmgr needs to assume.
>
>> I was actually thinking of
Robert Haas writes:
> On Fri, Dec 17, 2010 at 3:03 PM, Tom Lane wrote:
>> Yeah. I think that BM_UNLOGGED might be a poor choice for the flag name,
>> just because it overstates what the bufmgr needs to assume.
> I was actually thinking of adding BM_UNLOGGED even before this
> discussion, becaus
On Fri, Dec 17, 2010 at 3:03 PM, Tom Lane wrote:
> Heikki Linnakangas writes:
>> On 17.12.2010 21:32, Robert Haas wrote:
>>> I guess the question is whether it's right to conflate "table is
>>> unlogged" with "LSN is fake". It's not immediately obvious to me that
>>> those concepts are isomorphi
Heikki Linnakangas writes:
> On 17.12.2010 21:32, Robert Haas wrote:
>> I guess the question is whether it's right to conflate "table is
>> unlogged" with "LSN is fake". It's not immediately obvious to me that
>> those concepts are isomorphic, although though the reverse isn't
>> obvious to me ei
On 17.12.2010 21:32, Robert Haas wrote:
I guess the question is whether it's right to conflate "table is
unlogged" with "LSN is fake". It's not immediately obvious to me that
those concepts are isomorphic, although though the reverse isn't
obvious to me either.
The buffer manager only needs to
On Fri, Dec 17, 2010 at 2:31 PM, Tom Lane wrote:
> Robert Haas writes:
>> Another possibly-useful thing about mandating a full page header for
>> every page is that it might give us a way of avoiding unnecessary full
>> page writes. As I wrote previously:
>
> Could we do that via a bufmgr status
On Fri, Dec 17, 2010 at 2:22 PM, Tom Lane wrote:
> Heikki Linnakangas writes:
>> On 17.12.2010 21:07, Tom Lane wrote:
>>> IIUC, the problem is that the bufmgr might think that a GIST NSN is an
>>> LSN that should affect when to force out a dirty buffer? What if we
>>> taught it the difference?
Robert Haas writes:
> Another possibly-useful thing about mandating a full page header for
> every page is that it might give us a way of avoiding unnecessary full
> page writes. As I wrote previously:
Could we do that via a bufmgr status bit, instead? Heikki's idea has
the merit that it actual
On Fri, Dec 17, 2010 at 2:07 PM, Tom Lane wrote:
> Robert Haas writes:
> IIUC, the problem is that the bufmgr might think that a GIST NSN is an
> LSN that should affect when to force out a dirty buffer? What if we
> taught it the difference? We could for example dedicate a pd_flags
> bit to mar
Heikki Linnakangas writes:
> On 17.12.2010 21:07, Tom Lane wrote:
>> IIUC, the problem is that the bufmgr might think that a GIST NSN is an
>> LSN that should affect when to force out a dirty buffer? What if we
>> taught it the difference? We could for example dedicate a pd_flags
>> bit to marki
On 17.12.2010 21:07, Tom Lane wrote:
IIUC, the problem is that the bufmgr might think that a GIST NSN is an
LSN that should affect when to force out a dirty buffer? What if we
taught it the difference? We could for example dedicate a pd_flags
bit to marking pages whose pd_lsn isn't actually an
Robert Haas writes:
> Given the foregoing discussion, I see only two possible paths forward here.
> 1. Just decide that that unlogged tables can't have GIST indexes, at
> least until someone figures out a way to make it work. That's sort of
> an annoying limitation, but I think we could live wit
Given the foregoing discussion, I see only two possible paths forward here.
1. Just decide that that unlogged tables can't have GIST indexes, at
least until someone figures out a way to make it work. That's sort of
an annoying limitation, but I think we could live with it.
+1
In the small s
On Tue, Dec 14, 2010 at 5:14 PM, Robert Haas wrote:
> On Tue, Dec 14, 2010 at 4:55 PM, Tom Lane wrote:
>> Robert Haas writes:
>>> On Tue, Dec 14, 2010 at 4:24 PM, Heikki Linnakangas
>>> wrote:
Hmm, the first idea that comes to mind is to use a counter like the
GetXLogRecPtrForTemp() c
On Wed, Dec 15, 2010 at 4:06 AM, Simon Riggs wrote:
> On Sat, 2010-11-13 at 20:55 -0500, Robert Haas wrote:
>> I think that would be a recipe for bugs. Look at the three new macros
>> I introduced. If you keep relistemp around, then any code which
>> relies on it is likely testing for one of tho
On Wed, Dec 15, 2010 at 4:20 AM, Simon Riggs wrote:
> On Sat, 2010-11-13 at 19:16 -0500, Robert Haas wrote:
>
>> 3. The third patch (relax-sync-commit-v1) allows asynchronous commit
>> even when synchronous_commit=on if the transaction has not written
>> WAL. Of course, a read-only transaction wo
On Sat, 2010-11-13 at 19:16 -0500, Robert Haas wrote:
> 3. The third patch (relax-sync-commit-v1) allows asynchronous commit
> even when synchronous_commit=on if the transaction has not written
> WAL. Of course, a read-only transaction won't even have an XID and
> therefore won't need a commit re
On Sat, 2010-11-13 at 20:55 -0500, Robert Haas wrote:
> I think that would be a recipe for bugs. Look at the three new macros
> I introduced. If you keep relistemp around, then any code which
> relies on it is likely testing for one of those three things, or maybe
> even something subtly differen
On 14.11.2010 02:16, Robert Haas wrote:
3. The third patch (relax-sync-commit-v1) allows asynchronous commit
even when synchronous_commit=on if the transaction has not written
WAL. Of course, a read-only transaction won't even have an XID and
therefore won't need a commit record, so what this is
1 - 100 of 225 matches
Mail list logo
