How I suppose to handle a poller_ouput table is full

[Rno]

I got this error message this morning:

Code: Select all

2024/06/28 12:00:15 - SPINE: Poller[Main Poller] PID[59932] PT[140267541731072] ERROR: SQL Failed! Error:'1114', Message:'The table 'poller_output' is full', SQL Fragment:'INSERT INTO poller_output (local_data_id, rrd_name, time, output) VALUES (35533, 'ping', FROM_UNIXTIME(1719568805), '0.070'),(35536, 'proc', FROM_UNIXTIME(1719568805), '580'),(35539, '', FROM_UNIXTIME(1719568805), '1min:0.65 5min:0.60 10min:0.59'),(35542, 'users', FROM_UNIXTIME(1719568805), '0'),(35545, 'mem_buffers', FROM_UNIXTIME(1719568805), '228518260'),(35548, 'mem_swap', FROM_UNIXTIME(1719568805), '4194300'),(35551, 'ds_DNS_query_drop', FROM_UNIXTIME(1719568805), '415267'),(35554, 'ds_DNS_query_ipv4', FROM_UNIXTIME(1719568805), '-730103077'),(35557, 'ds_DNS_query_tcp', FROM_UNIXTIME(1719568805), '30366678'),(35560, 'ds_DNS_query_udp', FROM_UNIXTIME(1719568805), '-848981393'),(35563, '', FROM_UNIXTIME(1719568805), 'min:0 median:1 avg:1 max:6 iter:5'),(35566, '', FROM_UNIXTIME(1719568805), 'min:0 median:0 avg:0 max:3 iter:5') ON DUPLICATE KEY UPDATE output=VALUES(output)'

And on this table I have 3'810'974 records
What I can do ?
What is the table size limit ?

I'm not sure is a Cacti problem, because I'm doing some heavy redundancy test, and even if all the server (cacti, DB master, DB slave) have a redundant connection, it can be a cause of it.

Thanks for any input

[Rno]

Ok I find it
Problem with maxscale
Read on 1 server and read/write on the other

Should not be a problem anyway!

[TheWitness]

Yeah you have to exclude the poller_output table from replication and use the write only connection versus the readwrite split for poller. I've already crossed that bridge with MaxScale. You can still include it for replication but it's just wasted I/o cycles on the secondary server. The poller_output_boost table is okay for replication.