too many open connections, consider increasing the number of threads
Answered: amber (hollow confidence) — Paul Watson's terse pointer to tuning nfsd matches the kernel message origin, but the asker never confirmed. Other replies (keepalive, TIME_WAIT) address different connection-pileup scenarios.
Advisory only.
Posted in 2020
Neville Monteiro reported 'too many open connections, consider increasing the number of threads' every 2-3 seconds in /var/log/messages on RHEL 7.5, with load above 120 for 10-15 minutes. Paul Watson suggested tweaking nfsd if NFS runs. Art Kagel described dead client connections after slow IO, fixed by shortening OS keepalive and adding the keepalive option in sqlhosts. Vladimir Kolobrodov suggested checking TIME_WAIT. Unresolved.
Topics: Stored Procedures & SPL, Server Administration, Logging & Checkpoints
The following message is written repeatedly in the /var/log/messages file in the RHEL 7.5 (64 bit) system :-
"too many open connections, consider increasing the number of threads"
.
The above message is written to the /var/log/messages file after approx. every 2-3 seconds, so much so that the system becomes totally unresponsive. At this stage if you check the load of the system (using the "w -u" or "top" command), it displays that the system load to be more than 120. At this time, there are no error messages written in the online.log file of the databaseserver (only normal messages are written like "checkpoint completed: <number> seconds", "logical log <number> completed" etc ...). Even dbaccess command goes into a hang state.
Then, after approx. 10-15 minutes, the system load starts reducing and it comes back to less than 1. Then, the user are able to connect.
Just wanted to know as to what could be the problem and how do we set it right.
Thanks & Regards,
Neville Monteiro.
Sent with
ProtonMail
Secure Email.
#Informix
Running nfs ? Just tweak the nfsd Cheers Paul > The following message is written repeatedly in the /var/log/messages file > in the RHEL 7.5 (64 bit) system :- > "too many open connections, consider increasing the number of threads". > The above message is written to the /var/log/messages file after approx. > every 2-3 seconds, so much so that the system becomes totally > unresponsive. At this stage if you check the load of the system (using the > "w -u" or "top" command), it displays that the system load to be more than > 120. At this time, there are no error messages written in the online.log > file of the databaseserver (only normal messages are written like > "checkpoint completed: #Informix
Neville: I have a client that has seen a similar issue. They normally run around 4,000 active connections, but occasionally the number of connections starts to ramp up to over 30,000 bringing the system to its knees. The cause was an IO slow down that caused users get impatient waiting for a response from their applications and so to close their Windows based front-end applications and open a new copy. The windows apps were not shutting down the network connections to the database server. It was taking a long time for the OS to recognize that the clients were no longer attached to the other end of the connection and since the Informix server was not trying to write to those pipes it didn't know that the clients were gone. The solution was to adjust the keepalive setting in the OS to shorten the timeout and to add the keepalive option to the sqlhosts file lines. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves.
If you are doing a lot of (connect) disconnects on the system it's possible that you can run into issue related to TIME_WAIT. To confirm - look at the output of netstat -nta or, you could run something like netstat -nta | awk 'substr($1,1,3)=="tcp" {print $NF}' | sort | uniq -c There is nothing that can be done about it other than reduce frequency of disconnects, there are few workarounds including allowing reuse of the socket and reducing time spent in this state - but they are not without consequences. If you find that it's your problem - here are couple of references: TIME_WAIT and its design implications for protocols and scalable client server systems kernel: Possible SYN flooding on port #. Sending cookies. - Red Hat Customer Portal more can be found by searching for the TIME_WAIT ------------------------------ Vladimir Kolobrodov ------------------------------