Обсуждение: xlog viewer proposal

I'm developing the summer of code project to create a xlog viewer.<br />The tool we want to create is a DBA tool used
forinspect the xlog files, looking for some operations, statistcs of database usage and status of transactions. <br
/><br/>Some use cases:<br />* Some user made a mistake and commited it to the database. Now the DBA wants to find out
theexact xid to restore to a point in time where this mistake was yet to happen.<br /><br />* A long running
transactionchanged lots of data and we want to undo that. Show the most recent transactions, with their commit times
andtotal size of WAL for each transaction, so we can see the big transaction's xid. <br /><br />* How many write
transactionsper second am I getting?<br /><br />The idea I've been discussing with Simon Riggs is to create a set of
functionsthat can be called from within the database.<br />It seems that we would need to extract 2 relations from the
logfiles: transactions and xlog entries. <br />The functions that would extract the entries could easily look into any
xlogsegment passed as a parameter and return a relation containing these entries. But the functions needed to extract
thetransactions would need to look untill the last segment of the xlog to know the status of every transaction. <br
/><br/>The function to extract the xlog entries would read the XLogRecord structures (with all data associated), get a
TupleDescwith get_call_result_type() and return a tuple for each XLogRecord.<br /><br />Another problem is how to
presentthe specific data returned in each operation? <br />We can make functions to present this data in a humam
readableformat like: get_heap_data(), get_btree_data(), etc.<br />For example: SELECT get_btree_data(data) FROM
xlogviewer_file(xxx)WHERE operation = 'BTREE';<br /><br /> Other problem is how to sparete implict ABORTs from explict
ones?It seems that will be necessary to have separate functions for transaction info extraction. As I wrote above, this
functionswould have to read the xlogs all the way to the present moment to know which transactions are implicitly
aborted,and which ones are still active. <br /><br />This design gives lots of flexibility, we can use all the SQL
powerto query the xlog files. The one drawback is that you have to use a working backend to inspect the xlog, but in
caseswhere the database cluster is lost you could always use another cluster. It would be easy to create a wrapper
program(like createdb and createuser) to connect to a database and return the xlog info. <br /><br />Other advantage is
thepossibility of query remote xlogs trought a pg connection, it makes the remote managing easier and machine cluster
managingeasier to (don't have to make ssh accounts on all nodes or map a remote filesystem). <br /><br />Oracle has a
similartool called logminer (there goes an article about it <a
href="http://www.oracle.com/technology/oramag/oracle/05-jul/o45dba.html">http://www.oracle.com/technology/oramag/oracle/05-jul/o45dba.html
</a>).This postgresql xlogviewer would be also good for people migrating from oracle.<br /><br />Besides, if we create
asa separate program this would imply having useful functions related to backend data (the xlogs) not available to
otherbackend modules. It would be easier to create redundant code also. And I've read some emails about having already
duplicatecode for extracting text out of xlogs (the xlogs debug functions). <br /><br />Why a contrib module?<br
/>BecauseIt sounds safer to me to create a contrib module and it seems that I wont need to change the existing backend
code.<br />So all the code I'm planning to write is new, and I wont need changes in the backend. Another bonus is not
tobloat the backend and let this feature to be installed by those who really need it. Afterwards they can be integrated
inthe backend if needed. <br /><br />Given this design I would create some functions like (plus the data formating
functions):<br/><br />xlogviewer_file(char *file_path)<br />Return all entries in a given xlog segment<br /><br
/>xlogviewer_file_offset(char*file_path, uint32 offset) <br />Return all entries in a given xlog segment from an
offset<br/><br />xlogviewer_dir(char *directory_path)<br />Return all entries in all xlog segments inside a
directory<br/><br />xlogviewer_transactions(char *file_path)<br />Return all transactions from the directory containing
thesegment passed as parameter starting from this segment. <br /><br />One example of it's use:<br />SELECT * FROM
xlogviewer_file('00000001000000000000000F')xlog WHERE xlog.operation = 'BTREE';<br />The resultset would be something
like:<br/>xlog_record  |  previous_xlog_record  |  xid  |  operation  |  data <br
/>-------------+------------------------+-------+-------------+-------<br/> 0/00000220  | 0/000001F0             |
4    | BTREE       |<br /><br /><br />We could also query for a list of committed transactions:<br />SELECT<br />   
xlog.xid,count(1), sum(xlog.record_size)<br />FROM<br />    xlogviewer_file('00000001000000000000000F') xlog<br
/>WHERE<br/>    xlog.operation = 'XACT'<br />    <a href="http://xlog.info">xlog.info</a> = 'COMMIT'<br />GROUP BY<br
/>   xlog.xid<br /><br /><br clear="all" /><br />-- <br />Diogo Biazus - <a
href="mailto:diogob@gmail.com">diogob@gmail.com</a><br/>Móvel Consultoria<br /><a
href="http://www.movelinfo.com.br">http://www.movelinfo.com.br</a><br/><a href="http://www.postgresql.org.br">
http://www.postgresql.org.br</a>

"Diogo Biazus" <diogob@gmail.com> writes:
> The idea I've been discussing with Simon Riggs is to create a set of
> functions that can be called from within the database.

I'd question that at the very start.  I don't see any strong reason to
do it that way, and as you admit further down it'd make it impossible to
use the viewer to work on extracting data from a failed cluster; which
is, at least in my mind, one of the primary use-cases for the thing.
I would suggest building the xlog-reader as a separate program, not part
of the backend.  It would be useful to have options to (eg) translate
table OIDs to table names, which would require a connection to the
database, but this should be an *option* not an essential condition.
        regards, tom lane

On 6/22/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> it'd make it impossible to use the viewer to work
> on extracting data from a failed cluster; which is,
> at least in my mind, one of the primary use-cases
> for the thing.

While I too see this as something which could be used for this outside
the database, I don't think being able to access it from within the
system would be bad either.

> I would suggest building the xlog-reader as a separate
> program, not part of the backend.

I think it should certainly be able to run on it's own, but it
wouldn't be that hard to extend the functions so that they were usable
from within the database or vice-versa.

I just don't see it as mutually exclusive.  It would be relatively
easy to interface it from either side once the core functionality is
there.

-- 
Jonah H. Harris, Software Architect | phone: 732.331.1300
EnterpriseDB Corporation            | fax: 732.331.1301
33 Wood Ave S, 2nd Floor            | jharris@enterprisedb.com
Iselin, New Jersey 08830            | http://www.enterprisedb.com/

"Jonah H. Harris" <jonah.harris@gmail.com> writes:
> I think it should certainly be able to run on it's own, but it
> wouldn't be that hard to extend the functions so that they were usable
> from within the database or vice-versa.

Yes it would.  The most obvious point is that memory management and
error handling conventions inside the backend are quite different from
what you'd expect to employ in a standalone program.  Also the means
you'd use for consulting the system catalogs (in that option to provide
readable names for OIDs) are entirely different.

I think asking for support of both environments is a good way to ensure
that this summer project doesn't get finished :-(
        regards, tom lane

On 6/22/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Yes it would.  The most obvious point is that memory management and
> error handling conventions inside the backend are quite different from
> what you'd expect to employ in a standalone program.

No, this wouldn't really be that hard, especially if he created a few
macros to handle the differences.

> Also the means you'd use for consulting the system catalogs
> (in that option to provide readable names for OIDs) are entirely
> different.

Definitely correct there.  If it's designed well, it wouldn't be too
hard to use same-named functions for getting catalog information and
using ifdefs depending on how it's being used.

> I think asking for support of both environments is a good way to ensure
> that this summer project doesn't get finished :-(

I totally agree... I'm not suggesting that he does both for SoC.  Just
that if it is designed well from the start, it wouldn't be too hard to
add support for either command-line or contrib-ish usage.  The goal
was to start with xlogdump and enhance it, so starting with
command-line is probably the best anyway.

-- 
Jonah H. Harris, Software Architect | phone: 732.331.1300
EnterpriseDB Corporation            | fax: 732.331.1301
33 Wood Ave S, 2nd Floor            | jharris@enterprisedb.com
Iselin, New Jersey 08830            | http://www.enterprisedb.com/

"Jonah H. Harris" <jonah.harris@gmail.com> writes:
> On 6/22/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Yes it would.  The most obvious point is that memory management and
>> error handling conventions inside the backend are quite different from
>> what you'd expect to employ in a standalone program.

> No, this wouldn't really be that hard, especially if he created a few
> macros to handle the differences.

Jonah, I've been working with this system for years, and it's not that
easy to "handle the differences with a few macros".  I've seen people
try, repeatedly, and seen their code break repeatedly.  The relatively
small number of files that we use in both frontend and backend scenarios
are all extremely limited-in-function and tend to break easily.  I would
never try to build something nontrivial that would work both ways
... especially not on a very tight time budget.  Diogo will have enough
challenges delivering something useful that works in one environment.

> The goal
> was to start with xlogdump and enhance it, so starting with
> command-line is probably the best anyway.

Diogo, are you working from my old xlogdump hack?  If so what version?
I can send you the latest off-list.  I add stuff to it periodically when
I need it, and I don't think I've published it lately.
        regards, tom lane

Agree, the project must choose one path as the starting point. But the two options can be given in the long run.

I still think that as a starting point the functions inside the database are a good option.

The reasons are:
- using SQL to agregate and transform data in any way from the logs.
- it's easier for the DBA in the other use cases where the cluster is still active.
- give more flexibility for managing the xlogs remotely
- I think it's faster to implement and to have a working and usable tool.

And there is one option to minimize the problem in the failed cluster case: the wrapper program could give the option to initdb a temporary area when no connection is given, creating a backend just to analyze a set of xlogs.

After this summer project I could go on and try to use parts of this code to implement a realy standalone tool.

Other option is to start by the standalone tool and create a wrapper function inside postgresql that would just call this external program and extract data from the xlogs using this program's output (with some option to output all data in a CSV format).




--
Diogo Biazus - diogob@gmail.com
Móvel Consultoria
http://www.movelinfo.com.br
http://www.postgresql.org.br

Yup, I've got a version that was posted here some time ago. If you could send me the latest version I would be very glad.

--
Diogo Biazus - diogob@gmail.com
Móvel Consultoria
http://www.movelinfo.com.br
http://www.postgresql.org.br

On 6/22/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Jonah, I've been working with this system for years, and it's not that
> easy to "handle the differences with a few macros".

True, it is harder than just that.  I didn't mean to make light of it
at all, just that a good amount of design upfront would allow us to
support both.

> Diogo, are you working from my old xlogdump hack?  If so what version?
> I can send you the latest off-list.  I add stuff to it periodically when
> I need it, and I don't think I've published it lately.

As it's a pretty cool and useful utility, could you publish it on-list
again some time soon?  It would be nice to pull some stuff from it for
the pg_resetxlog -f stuff... rather than having to start from scratch
copying & changing some code from xlog.c.

-- 
Jonah H. Harris, Software Architect | phone: 732.331.1300
EnterpriseDB Corporation            | fax: 732.331.1301
33 Wood Ave S, 2nd Floor            | jharris@enterprisedb.com
Iselin, New Jersey 08830            | http://www.enterprisedb.com/

On Thu, 2006-06-22 at 14:57 -0300, Diogo Biazus wrote:
> Agree, the project must choose one path as the starting point. But the
> two options can be given in the long run.

I'm acting as Diogo's mentor for the SoC, so I'm trying to let Diogo
discuss his ideas in the community manner without too much steering.

Diogo's ideas are interesting - they aren't the way I would have done it
either, but that doesn't mean we shouldn't consider this alternative
approach.

> I still think that as a starting point the functions inside the
> database are a good option.

Yes, if we use SRF functions for this, ISTM they are the best place for
them. 

> The reasons are: 
> - using SQL to agregate and transform data in any way from the logs.

That is a major point here. If the xlogdump is purely a stand-alone
program that it will be much less functionally rich and as Tom mentions,
there are other reasons for having access to a server.

> - it's easier for the DBA in the other use cases where the cluster is
> still active. 

Good point.

> - give more flexibility for managing the xlogs remotely

Not sure what you mean.

> - I think it's faster to implement and to have a working and usable
> tool.

Why do you think that? It sounds like you've got more work since you
effectively need to rewrite the _desc routines.

> And there is one option to minimize the problem in the failed cluster
> case: the wrapper program could give the option to initdb a temporary
> area when no connection is given, creating a backend just to analyze a
> set of xlogs. 

It seems a reasonable assumption that someone reading PostgreSQL logs
would have access to another PostgreSQL cluster. It obviously needs to
work when the server that originated the logs is unavailable, but that
does not mean that all PostgreSQL systems are unavailable. There's no
need to try to wrap initdb - just note that people would have to have
access to a PostgreSQL system.

> Other option is to start by the standalone tool and create a wrapper
> function inside postgresql that would just call this external program
> and extract data from the xlogs using this program's output (with some
> option to output all data in a CSV format). 

I think this idea is a good one, but we must also consider whether is
can be done effectively within the time available. Is this: can do now
or want to do in future?

The alternative of reinforcing xlogdump needs to be considered more
fully now and quickly, so coding can begin as soon as possible. 
- Diogo: what additional things can you make xlogdump do?
- Tom: can you say more about what you'd like to see from a tool, to
help Diogo determine the best way forward. What value can he add if you
have already written the tool?


Some other considerations:
The biggest difficulty is finding "loser transactions" - ones that have
not yet committed by the end of the log. You need to do this in both
cases if you want to allow transaction state to be determined precisely
for 100% of transactions; otherwise you might have to have an Unknown
transaction state in addition to the others.

What nobody has mentioned is that connecting to a db to lookup table
names from OIDs is only possible if that db knows about the set of
tables the log files refer to. How would we be certain that the
OID-to-tablename match would be a reliable one?

-- Simon Riggs              EnterpriseDB   http://www.enterprisedb.com

Good question, It seems to me that the only case where this have a trivial aswer is if your inside the backend querying the current xlog directory.
I'm still thinking about the solution for the other cases (inside or outside the backend).

--
Diogo Biazus - diogob@gmail.com
Móvel Consultoria
http://www.movelinfo.com.br
http://www.postgresql.org.br

--
Diogo Biazus - diogob@gmail.com
Móvel Consultoria
http://www.movelinfo.com.br
http://www.postgresql.org.br

On Fri, 2006-06-23 at 11:03 -0300, Diogo Biazus wrote:
> On 6/23/06, Diogo Biazus <diogob@gmail.com> wrote:


>         On 6/23/06, Simon Riggs <simon@2ndquadrant.com> wrote:
>                 > - give more flexibility for managing the xlogs
>                 remotely
>                 
>                 Not sure what you mean.
> 
> I can connect to the server if I want to query xlogs in a remote
> machine.
> If i depend on a standalone tool that reads the filesystem I'll need
> some kind of ssh access.
> Sometimes the DBA needs the info in the xlogs but doesn't have/want
> access to the SO.
> It's also easier to create a script that collect statistics about
> xlogs in machiune clusters. 

OK, security is an interesting one.

--  Simon Riggs              EnterpriseDB   http://www.enterprisedb.com

On Fri, 2006-06-23 at 10:59 -0300, Diogo Biazus wrote:

> On 6/23/06, Simon Riggs <simon@2ndquadrant.com> wrote:
>         > - give more flexibility for managing the xlogs remotely
>         
>         Not sure what you mean.
>         
>         > - I think it's faster to implement and to have a working and
>         usable
>         > tool.
>         
>         Why do you think that? It sounds like you've got more work
>         since you 
>         effectively need to rewrite the _desc routines.
> 
> Yes, but I don't need to worry with program output, and I have the
> backend's memory management and error handling.

I'd suggest doing a quick prototype to allow us to evaluate which
architecture would be preferable.

I'm torn between the good-idea and the safe-minimal-but-definitely in
8.2 option.

--  Simon Riggs              EnterpriseDB   http://www.enterprisedb.com

Alright, I'm working on a fast prototype using the SRF.




--
Diogo Biazus - diogob@gmail.com
Móvel Consultoria
http://www.movelinfo.com.br
http://www.postgresql.org.br