Обсуждение: network_ops in 7.0 and pg_dump question
Hi,
what's happen with network_ops in current CVS ?
I just synced sources and couldn't load dump from 6.5.3 -
problem occures on
CREATE INDEX "face_key" on "face" using btree ( "eid" "int4_ops", "ip" "network_ops" );
The message I got:
CREATE
ERROR: DefineIndex: network_ops class not found
Table face:
election=# \d face Table "face"Attribute | Type | Modifier
-----------+------------+----------eid | integer | ip | inet | vdate | datetime | ftrs
|smallint[] |
Also, does new pg_dump is aware about order of defining of function
and tables, when function is used in CREATE TABLE, for example:
CREATE TABLE "applicant" ( "candx" int2 DEFAULT next_applicant ( ) NOT NULL, "candidate" text,
"candt"int2, "img" text);
but function next_applicant() is dumped in 6.5.3 after CREATE TABLE
and this cause an error. I had manually edit dump file to reverse order :-)
Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, sci.researcher, hostmaster of AstroNet,
Sternberg Astronomical Institute, Moscow University (Russia)
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(095)939-16-83, +007(095)939-23-83
> Hi, > > what's happen with network_ops in current CVS ? > I just synced sources and couldn't load dump from 6.5.3 - > problem occures on > CREATE INDEX "face_key" on "face" using btree ( "eid" "int4_ops", "ip" "network_ops" ); > > The message I got: > CREATE > ERROR: DefineIndex: network_ops class not found > Oops, my fault. There was some confusing links in the catalog for the ip/cidr types. They pointed to the same *ops, which made the table non-unique, so the cache would grab a random matching entry. The new system has separate *ops for each type. We were basically using the cache on a non-unique entry. We would grab the first match. The new code uses the same underlying functions, but moves the duplication down one level. Now, how to convert these? Not supplying the ops works fine, but pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops, we should just remove that from being passed to the backend for a few releases. Comments? -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Now, how to convert these? Not supplying the ops works fine, but
> pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops,
> we should just remove that from being passed to the backend for a few
> releases. Comments?
Ugly, but probably the best stopgap for backwards compatibility ...
at least I can't think of a better answer, since we have no way to
change what 6.5 pg_dump will dump.
You're only going to suppress "network_ops" if it appears in the
ops position of a CREATE INDEX, right? Don't want to stop people
from using the name for fields and so on.
regards, tom lane
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Now, how to convert these? Not supplying the ops works fine, but
> pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops,
> we should just remove that from being passed to the backend for a few
> releases. Comments?
Actually, rather than hacking gram.y, it seems like it would be cleaner
to put the kluge in whatever part of the parser looks up the ops name.
Of course a kluge is a kluge no matter what...
regards, tom lane
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > > Now, how to convert these? Not supplying the ops works fine, but > > pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops, > > we should just remove that from being passed to the backend for a few > > releases. Comments? > > Ugly, but probably the best stopgap for backwards compatibility ... > at least I can't think of a better answer, since we have no way to > change what 6.5 pg_dump will dump. > > You're only going to suppress "network_ops" if it appears in the > ops position of a CREATE INDEX, right? Don't want to stop people > from using the name for fields and so on. No, just at that part in the grammar. -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > > Now, how to convert these? Not supplying the ops works fine, but > > pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops, > > we should just remove that from being passed to the backend for a few > > releases. Comments? > > Actually, rather than hacking gram.y, it seems like it would be cleaner > to put the kluge in whatever part of the parser looks up the ops name. > > Of course a kluge is a kluge no matter what... I like it in gram.y because it is more visible there and easier to remove later. -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Thanks,
creation of index works now. But what about pg_dump ?
I still have to edit manually dump file.
look to excerption from dump file:
CREATE TABLE "applicant" ( "candx" int2 DEFAULT next_applicant() NOT NULL, "candidate" text, "candt"
int2, "img" text
);
This fails because function next_applicant dumps later !
Here is a psql output:
You are now connected as new user megera.
ERROR: Relation 'applicant' does not exist
invalid command \N
invalid command \N
invalid command \N
invalid command \N
invalid command \.
ERROR: parser: parse error at or near "2"
invalid command \.
ERROR: parser: parse error at or near "1"
invalid command \.
ERROR: parser: parse error at or near "1"
invalid command \.
ERROR: parser: parse error at or near "1"
invalid command \.
ERROR: parser: parse error at or near "24"
invalid command \.
ERROR: parser: parse error at or near "24"
CREATE
CREATE
Hmm, error diagnostics still not very informative :-)
Regards,
Oleg
On Mon, 7 Feb 2000, Bruce Momjian wrote:
> Date: Mon, 7 Feb 2000 19:03:29 -0500 (EST)
> From: Bruce Momjian <pgman@candle.pha.pa.us>
> To: Tom Lane <tgl@sss.pgh.pa.us>
> Cc: Oleg Bartunov <oleg@sai.msu.su>, hackers@postgreSQL.org
> Subject: Re: [HACKERS] network_ops in 7.0 and pg_dump question
>
> > Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > > Now, how to convert these? Not supplying the ops works fine, but
> > > pg_dump supplies the ops. Maybe in gram.y, if they supply network_ops,
> > > we should just remove that from being passed to the backend for a few
> > > releases. Comments?
> >
> > Ugly, but probably the best stopgap for backwards compatibility ...
> > at least I can't think of a better answer, since we have no way to
> > change what 6.5 pg_dump will dump.
> >
> > You're only going to suppress "network_ops" if it appears in the
> > ops position of a CREATE INDEX, right? Don't want to stop people
> > from using the name for fields and so on.
>
> No, just at that part in the grammar.
>
> --
> Bruce Momjian | http://www.op.net/~candle
> pgman@candle.pha.pa.us | (610) 853-3000
> + If your life is a hard drive, | 830 Blythe Avenue
> + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
>
> ************
>
_____________________________________________________________
Oleg Bartunov, sci.researcher, hostmaster of AstroNet,
Sternberg Astronomical Institute, Moscow University (Russia)
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(095)939-16-83, +007(095)939-23-83
Oleg Bartunov <oleg@sai.msu.su> writes:
> creation of index works now. But what about pg_dump ?
> I still have to edit manually dump file.
> look to excerption from dump file:
> CREATE TABLE "applicant" (
> "candx" int2 DEFAULT next_applicant() NOT NULL,
> "candidate" text,
> "candt" int2,
> "img" text
> );
> This fails because function next_applicant dumps later !
Yeah, it's a known bug. We can't just dump the functions first,
though, can we? I'm not sure how carefully function definitions
get examined by CREATE FUNCTION.
The simplest real solution I've heard so far is to dump database objects
in order by OID rather than doing it strictly by type.
Is anyone working on this, or does anyone want to? I haven't looked at
pg_dump in a while, but I know some other folks have been hacking it
recently.
regards, tom lane
> Yeah, it's a known bug. We can't just dump the functions first, > though, can we? I'm not sure how carefully function definitions > get examined by CREATE FUNCTION. > > The simplest real solution I've heard so far is to dump database objects > in order by OID rather than doing it strictly by type. > > Is anyone working on this, or does anyone want to? I haven't looked at > pg_dump in a while, but I know some other folks have been hacking it > recently. I thought Peter E. was thinking about it. -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Tom Lane wrote: > > The simplest real solution I've heard so far is to dump database objects > in order by OID rather than doing it strictly by type. > > Is anyone working on this, or does anyone want to? I haven't looked at > pg_dump in a while, but I know some other folks have been hacking it > recently. I'll take a stab at it, if Peter E. isn't already doing it. -- Mark Hollomon mhh@nortelnetworks.com ESN 451-9008 (302)454-9008
At 02:01 PM 2/8/00 -0500, Mark Hollomon wrote: >Tom Lane wrote: >> >> The simplest real solution I've heard so far is to dump database objects >> in order by OID rather than doing it strictly by type. >> >> Is anyone working on this, or does anyone want to? I haven't looked at >> pg_dump in a while, but I know some other folks have been hacking it >> recently. > >I'll take a stab at it, if Peter E. isn't already doing it. You might want to e-mail Jan and/or Steve Szabo, who've been working on dumping referential integrity stuff. Because tables can mutally refer to each other, constraint dumping won't be done until data is dumped, so the data will be loaded first when someone recreates the database from the dump. I was busy over the weekend working the MATCH <unspecified> and the semantics of referential integrity actions so mostly ignored the e-mails they traded on the subject - you'll need to get details from them. You need to make sure whatever you do doesn't break whatever they've done or are doing... - Don Baccus, Portland OR <dhogaza@pacifier.com> Nature photos, on-line guides, Pacific Northwest Rare Bird Alert Serviceand other goodies at http://donb.photo.net.
> The simplest real solution I've heard so far is to dump database objects > in order by OID rather than doing it strictly by type. > > Is anyone working on this, or does anyone want to? I haven't looked at > pg_dump in a while, but I know some other folks have been hacking it > recently. Dumping by Oid or building up a framework of dependencies, these where the options. Don't forget, SQL language functions are (in contrast to procedural ones) parsed at CREATE time. So any operator, aggregate or table you use inside must exist. And they can be used in turn in many places, so it isn't simple at all. I think finally pg_dump must scan the entire schema two times, first to get all the Oid's, second to dumpall the objects. AFAIK, nobody is working on it. And starting on it right now seems a little late to make it until BETA. Jan -- #======================================================================# # It's easier to get forgiveness for being wrong than for being right. # # Let's break this rule - forgive me. # #========================================= wieck@debis.com (Jan Wieck) #
> Tom Lane wrote:
> >
> > The simplest real solution I've heard so far is to dump database objects
> > in order by OID rather than doing it strictly by type.
Hmm. Now if my OO stuff was working I guess pg_dump could be implemented
as...
List<PGObject*> dblist = pgselect("SELECT ** from object order by oid");
while (dblist.begin(); !dblist.atEnd(); dblist++) {dblist.obj().dump();
On 2000-02-08, Tom Lane mentioned: > The simplest real solution I've heard so far is to dump database objects > in order by OID rather than doing it strictly by type. AFAIR, it was your idea ... ;) > > Is anyone working on this, or does anyone want to? I haven't looked at > pg_dump in a while, but I know some other folks have been hacking it > recently. I might have been putting out remarks to that end once in a while, and I'm still interested in it, but it would be a more extensive project, like the psql revision, because pg_dump needs a lot of love as it stands. (I think there are some parts still in it that allow you to dump PostQUEL.) The problem with a pure oid-based ordering concept is that (as you yourself pointed out) it won't work if you alter some object in question after creation. The obvious case would be an alter function (to be implemented), but another case is (probably) alter column set default (is implemented). What I'd like to do first is to draw up some (semi-)formal (dependency-based) concept on paper and either verify it or come to the conclusion that it will never work and then give up in disgust. ;) No, seriously, I suppose I'll bring this up again in a couple of months when we're ready for it. Any collaborators are welcome of course. -- Peter Eisentraut Sernanders väg 10:115 peter_e@gmx.net 75262 Uppsala http://yi.org/peter-e/ Sweden
Peter Eisentraut <peter_e@gmx.net> writes:
> The problem with a pure oid-based ordering concept is that (as you
> yourself pointed out) it won't work if you alter some object in question
> after creation. The obvious case would be an alter function (to be
> implemented), but another case is (probably) alter column set default (is
> implemented).
Right; a genuine dependency analysis would be better. Also a lot more
painful to implement.
As you say, pg_dump could do with a wholesale rewrite, and maybe that
would be a good time to look at the dependency-based approach. In the
meantime, I think dumping in OID order would fix 90% of the problem for
10% of the work...
regards, tom lane