OK, two problems.
Problem number one seems to have been introduced with the LocalSocket option.
In fact, that config value is ignored in favor of its key, the literal
"LocalSocket". So the plugin creates the socket as "LocalSocket", connects to
the pdns control socket and sends the data. When the pdns control socket
attempts to send data back, it finds no socket called "LocalSocket" as it's in
a different working directory. Of course, this shouldn't cause collectd to
hang. That's the second problem. The plugin's socket has no recv timeout.
Since it was able to connect and send data successfully, it waits indefinitely
for a response that's not coming.
Both problems are fixed in the attached patch, which I made against the squeeze
debian source. I've tested the resulting package successfully in a lenny
environment, both with and without a chrooted pdns_recursor.
Luke Heberling
--- collectd-4.6.3.broken/src/powerdns.c 2009-06-02 02:17:47.000000000 -0700
+++ collectd-4.6.3/src/powerdns.c 2009-07-27 16:44:32.000000000 -0700
@@ -380,6 +380,16 @@ static int powerdns_get_data_dgram (list
break;
}
+ struct timeval timeout;
+ timeout.tv_sec=5;
+ timeout.tv_usec=0;
+ status = setsockopt (sd, SOL_SOCKET, SO_RCVTIMEO, &timeout, sizeof (timeout));
+ if (status != 0)
+ {
+ FUNC_ERROR ("setsockopt");
+ break;
+ }
+
status = connect (sd, (struct sockaddr *) &item->sockaddr,
sizeof (item->sockaddr));
if (status != 0)
@@ -911,11 +921,18 @@ static int powerdns_config (oconfig_item
powerdns_config_add_server (option);
else if (strcasecmp ("LocalSocket", option->key) == 0)
{
- char *temp = strdup (option->key);
- if (temp == NULL)
- return (1);
- sfree (local_sockpath);
- local_sockpath = temp;
+ if ((option->values_num != 1) || (option->values[0].type != OCONFIG_TYPE_STRING))
+ {
+ WARNING ("powerdns plugin: `%s' needs exactly one string argument.", ci->key);
+ }
+ else
+ {
+ char *temp = strdup (option->values[0].value.string);
+ if (temp == NULL)
+ return (1);
+ sfree (local_sockpath);
+ local_sockpath = temp;
+ }
}
else
{
