Subject: Re: GIN pending list clean up exposure to SQL
To: Fujii Masao <masao.fujii@gmail.com>
References: 
 <CAMkU=1x8zFkpfnozXyt40zmR3Ub_kHu58LtRmwHUKRgQss7=iQ@mail.gmail.com>
 <CAJGNTeNVBMDzxwi=QaJWbeVequf1oENF9GbFaRMfny9H6ic_Ng@mail.gmail.com>
 <CAMkU=1xdv5n8w2-n4j_ysXy6eRd69XhWiGrdxQbqDjWEpBavww@mail.gmail.com>
 <CAMkU=1wEzVkxDJSQe5gbwb7AQ7zMi3D7mr-XMQBoyPzUnzpSpQ@mail.gmail.com>
 <56924D75.7060801@dalibo.com>
 <CAMkU=1w0z86n=+N1yFfKx4-wrO1m6M9XvEFBhV6TQmV4sK1waw@mail.gmail.com>
 <569975CD.3070103@dalibo.com>
 <CAHGQGwH=M1BAEJPQdpJjCneqwg8Xa+P8SB+ZsvhVwH6gL2J46w@mail.gmail.com>
Cc: Jeff Janes <jeff.janes@gmail.com>,
 Jaime Casanova <jaime.casanova@2ndquadrant.com>,
 pgsql-hackers <pgsql-hackers@postgresql.org>
From: Julien Rouhaud <julien.rouhaud@dalibo.com>
Message-ID: <56A28AAA.6050502@dalibo.com>
Date: Fri, 22 Jan 2016 21:01:46 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.5.1
MIME-Version: 1.0
In-Reply-To: 
 <CAHGQGwH=M1BAEJPQdpJjCneqwg8Xa+P8SB+ZsvhVwH6gL2J46w@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 8bit
Precedence: bulk
Sender: pgsql-hackers-owner@postgresql.org

On 20/01/2016 15:17, Fujii Masao wrote:
> 
> When I compiled the PostgreSQL with the patch, I got the following error.
> ISTM that the inclusion of pg_am.h header file is missing in ginfast.c.
> 
> ginfast.c: In function 'gin_clean_pending_list':
> ginfast.c:980: error: 'GIN_AM_OID' undeclared (first use in this function)
> ginfast.c:980: error: (Each undeclared identifier is reported only once
> ginfast.c:980: error: for each function it appears in.)
> 
> gin_clean_pending_list() must check whether the server is in recovery or not.
> If it's in recovery, the function must exit with an error. That is, IMO,
> something like the following check must be added.
> 
>          if (RecoveryInProgress())
>                  ereport(ERROR,
> 
> (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
>                                   errmsg("recovery is in progress"),
>                                   errhint("GIN pending list cannot be
> cleaned up during recovery.")));
> 
> It's better to make gin_clean_pending_list() check whether the target index
> is temporary index of other sessions or not, like pgstatginindex() does.
> 
> +    Relation    indexRel = index_open(indexoid, AccessShareLock);
> 
> ISTM that AccessShareLock is not safe when updating the pending list and
> GIN index main structure. Probaby we should use RowExclusiveLock?
> 

This lock should be safe, as pointed by Alvaro and Jaime earlier in this
thread
(http://www.postgresql.org/message-id/20151119161846.GK614468@alvherre.pgsql),
as ginInsertCleanup() can be called concurrently.

Also, since 38710a374ea the ginInsertCleanup() call must be fixed:

-   ginInsertCleanup(&ginstate, false, true, &stats);
+   ginInsertCleanup(&ginstate, true, &stats);


Regards.

-- 
Julien Rouhaud
http://dalibo.com - http://dalibo.org


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