Thread: Incomplete Startup Packet

Some time ago, I accidentally did a kill -9 on the postmaster (yes, I
know, I know), when trying to kill -9 one of the child processes (er,
yeah, probably bad too). This turned out to be pretty bad for us. It put
the database in a bad state. I had to run some kind of hacky command (I
don't recall which one) to even get postgres to start up again. Since
then, the log file is littered with:

LOG:  incomplete startup packet

I am ok with the fact that the abrupt killing of the postmaster may have
corrupted some data. It is not a mission critical data we're talking
about. But I'm left with some questions -

Is my database hosed? Does this necessitate a full reinstall of
postgres? While not mission critical data, there is a lot of it, and
many dbs in the cluster which would mean hours of data loading. (The
database seems to be functioning just fine, but I seem to recall reading
that a reinstall is recommended, though I forget why)

- Mott

As a side question, probably unrelated -- i understand that kill -9
postmaster is bad, but how bout killing a child process (a client)? I've
noticed that if you kill a child process, it seems to kill all child
processes and reboot (like a SIGUP?) [I was doing this in order to kill
a hanging transaction.]

Mott Leroy <mott@acadaca.com> writes:
> Some time ago, I accidentally did a kill -9 on the postmaster (yes, I
> know, I know), when trying to kill -9 one of the child processes (er,
> yeah, probably bad too). This turned out to be pretty bad for us. It put
> the database in a bad state. I had to run some kind of hacky command (I
> don't recall which one) to even get postgres to start up again. Since
> then, the log file is littered with:

> LOG:  incomplete startup packet

It's impossible to believe that that's even slightly related.
"Incomplete startup packet" means that you've got broken client-side
software, or perhaps that something is portscanning you.  You sure you
weren't seeing those beforehand?

            regards, tom lane

Tom Lane wrote:

> It's impossible to believe that that's even slightly related.
> "Incomplete startup packet" means that you've got broken client-side
> software, or perhaps that something is portscanning you.  You sure you
> weren't seeing those beforehand?
>

I'm not certain I wasn't seeing those beforehand, no. I suppose I jumped
to conclusions a bit. Should I be concerned about these "incomplete
startup packet" errors? It's unlikely that its a portscan, since the db
server is invisible to the outside world. How can I go about tracking
down the source of these? My db clients are JDBC connections from web
applications.

Mott Leroy <mott@acadaca.com> writes:
> How can I go about tracking down the source of these?

I think if you turn on log_connections, you'll at least get log entries
showing what machine(s) the bad connection attempts are coming from.
Not sure if that'll be enough for you.

> My db clients are JDBC connections from web applications.

You might try asking about it on the pgsql-jdbc list; perhaps there's a
known issue of this kind.

            regards, tom lane

Markus Wollny wrote:

> We're getting "incomplete startup packet" messages in our logfiles due
> to some sort of system probe run by our service provider which checks if
> PG is still running. In our case they're harmless of course. Are you
> sure that you're not running something along those lines, too?

Ah, in fact, that is the case. We have Nagios running which checks to
see if postgres is still up. It very well may be that this is the cause
of the messages. Thank you.

Mott