[GENERAL] Startup messages for socket protocol
Hi.
I've been trying to get a connection working to a PostgreSQL server
through the socket-level protocol. I've therefore been looking at
fe-connect.c and fe-protocol3.c in the src/interfaces/libpq folder.
Reading those sources, I understood, that the startup message should
begin with a request from the client with the protocol number.
In fe-protocol3.c, build_startup_message():
/* Protocol version comes first. */
if (packet)
{
ProtocolVersion pv = htonl(conn-pversion);
memcpy(packet + packet_len, pv, sizeof(ProtocolVersion));
}
packet_len += sizeof(ProtocolVersion);
However, when I try to send this as the first thing, I get disconnected
by the server. Reading what the psql program does, I first get an 8 byte
message containing this:
\000\000\000\008\004\210\022/
This seems to work, but I'm at a loss pinpointing in the libpq source
code where that would fit in the protocol.
Enlightenment would be very welcome.
--
Guillaume Yziquel
--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
Re: [GENERAL] Startup messages for socket protocol
Le Thursday 17 Mar 2011 à 10:48:50 (-0400), Tom Lane a écrit :
Guillaume Yziquel guillaume.yziq...@citycable.ch writes:
However, when I try to send this as the first thing, I get disconnected
by the server. Reading what the psql program does, I first get an 8 byte
message containing this:
\000\000\000\008\004\210\022/
This seems to work, but I'm at a loss pinpointing in the libpq source
code where that would fit in the protocol.
[ scratches head... ] You should be getting either an ErrorResponse
message or some AuthenticationXXX variant, and both of those would start
with an ASCII character ('E' or 'R').
For now, when sending \000\003\000\000 and only this, the server seems
to disconnect. The recv() call on the socket returns 0, which should
mean that the server has dropped the connection.
I'm not sure what the above could
be, unless maybe you're somehow triggering an SSL startup handshake ---
For now, I simply have a INET socket. Just sending \000\003\000\000. No
SSL handshake.
but the first returned byte should be an 'S' in that case. Are you sure
you've correctly identified what is payload data, versus what's TCP
overhead or something like that?
Not sure how to identify that. What I identified was 'cannot read and
server disconnects client'. When I send \000\000\000\008\004\210\022/
(which is what the psql sent me when I looked it up), I get a N, and
it waits for further data (i.e. not disconnected yet...). Then I get
disconnected after some timeout, I guess.
It might also be enlightening to look into the server's log, especially
if you were to crank log_min_messages way up so it logs debug stuff.
Not so familiar as to where to look, except for the file in
/var/log/postgresql:
2011-03-17 14:41:34 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 14:42:12 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 14:42:21 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 14:43:46 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 14:45:01 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 14:46:38 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 15:08:04 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 15:33:08 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 15:35:36 CET LOG: longueur invalide du paquet de d?marrage
2011-03-17 16:01:16 CET LOG: longueur invalide du paquet de d?marrage
In english: invalid length for startup packet.
regards, tom lane
What should a typical startup packet be? psql confuses me more than it
helps me.
--
Guillaume Yziquel
--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
Re: [GENERAL] Startup messages for socket protocol
Le Thursday 17 Mar 2011 à 16:08:55 (+0100), Guillaume Yziquel a écrit :
Le Thursday 17 Mar 2011 à 10:48:50 (-0400), Tom Lane a écrit :
Guillaume Yziquel guillaume.yziq...@citycable.ch writes:
For now, when sending \000\003\000\000 and only this, the server seems
to disconnect. The recv() call on the socket returns 0, which should
mean that the server has dropped the connection.
Got it:
Sent: \000\000\000\022\000\003\000\000user\000yziquel\000\000
Read from socket:
R\000\000\000\b\000\000\000\000S\000\000\000\025client_encoding\000UTF8\000S\000\000\000\023DateStyle\000ISO,
DMY\000S\000\000\000\025integer_datetimes\000on\000S\000\000\000\027IntervalStyle\000postgres\000S\000\000\000\021is_superuser\000off\000S\000\000\000\025server_encoding\000UTF8\000S\000\000\000\025server_version\0008.4.7\000S\000\000\000\session_authorization\000yziquel\000S\000\000\000$standard_conforming_strings\000off\000S\000\000\000\023TimeZone\000localtime\000K\000\000\000\012\000\000|\197{\177\235?Z\000\000\000\005I
Needed to prepend the length of the packet.
Didn't appear very clearly in the docs. But this link got me more info:
http://blog.endpoint.com/2010/05/finding-postgresql-version-without.html
Thanks for your time.
--
Guillaume Yziquel
--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
[GENERAL] PQntuples returns an int.
Hello. I've been looking at the PQntuples function, in section 29.3.2. 'Retrieving query result information' of http://www.postgresql.org/docs/8.2/interactive/libpq-exec.html The declaration is: int PQntuples(const PGresult *res); I'm wondering: why the 'int'? why not a 'long int', or an 'unsigned int', or whatever? I'm asking this because I'm wondering whether you could have more tuples in the result of a query than could be indexed by the 'int' type. May this happen? All the best, -- Guillaume Yziquel http://yziquel.homelinux.org/ -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
Re: [GENERAL] PQntuples returns an int.
Greg Sabino Mullane a écrit : -BEGIN PGP SIGNED MESSAGE- Hash: RIPEMD160 I'm asking this because I'm wondering whether you could have more tuples in the result of a query than could be indexed by the 'int' type. May this happen? Yes, but it's a low concern as many other things would break before you got to that point. See: http://groups.google.com/group/pgsql.general/browse_thread/thread/462c454345221267/f61926526b539036 or: http://5z8.info/super-nsfw_cyd Thank you. Indeed, I perfectly do believe you when you state that many things would break before. It's much more a typing issue in a strongly typed language (OCaml). Thanks again. Guillaume Yziquel. -- Guillaume Yziquel http://yziquel.homelinux.org/ -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
