Unanswered question

How to solve NL-NETWORK-01 error during a load test ?

During the execution of a limit load test for a client, everything happens correctly at the beginning, but NL-NETWORK-01 errors suddenly appear after a certain heavy load.

Error Code: NL-NETWORK-01
Message: Network error: An IO error occurred sending the request.
Details : java.io.EOFException: HttpConnectionOverHTTP@69ed9b49(l:/10.178.8.127:35252 <-> r:wamwg-rec.recgroupement.systeme-u.fr/10.178.xxx.xx:443,closed=false)[HttpChannelOverHTTP@174acb75(exchange=HttpExchange@2aa0a348 req=TERMINATED/null@null res=PENDING/null@null)[send=HttpSenderOverHTTP@3a09a766(req=QUEUED,snd=COMPLETED,failure=null)[HttpGenerator@3f3b44bd{s=START}],recv=HttpReceiverOverHTTP@7eeb314c(rsp=IDLE,failure=null)[HttpParser{s=CLOSED,0 of 0}]]]

In a previous post on the forum dating from 2016 (http://answers.neotys.com/questions/1144936-facing-network-error-io-error-occurred-sending-request-error-single-user), the support indicated that : "It's a kind of connection reset. If it happens during a test you should check any network devices between your load generator and your application. It can be an issue due to firewall, proxy, load balancer or even the application under load."
But I would have liked to know if the analysis track was always similar or if new tracks could be evoked ? Also, I wanted to know if the cause could come from the Neoload injection infrastructure ?
I am in version 6.6.0 of Neoload and for information, I applied the setting recommended by Neoload for tests with high load (run WindowsNetworkTuning.reg and restart, etc ...). Also, a test with 4 injectors instead of 3 has to generate the same errors after exactly the same time.
Would you have a resolution track for this problem ?
Thank you very much

Yohann H.
Yohann H.

Yohann H.

Level
0
55 / 100
points

Answers

Do you have monitoring in place? the fact that it occurs at the same place each time and you have added more LGs without improvement suggests the issue is on the application side. I would suggest monitoring of the app server and database may reveal a pattern.

Yohann H.
Yohann H.

Yohann H.

Level
0
55 / 100
points

Thank you for your answer Neil. Yes, the entire application infrastructure is monitored and unfortunately, no CPU or machine memory issues or JVM issues (memory, threads, GC) have been identified. It is an authentication infrastructure that is tested with Directory Services so without a database, but I really wanted to know if I could remove the track of the injection infrastructure to focus my analysis on the tested environment (network, firewall, etc ...). Thank you.

The test infrastructure should be monitored by default, if there are no clues there then we may struggle to find the cause. It can’t be firewalls as initially the test works so connection wise, that sounds okay. You could look in the Neoload logs?

@Yohann. This kind of errors are difficult to track down as it can be any network devices even for example an anti-virus.

As a guide line first make sure that you do not have any anti-virus on your LG machines. You can use a Linux machine to make sure. Then the goal would be to use an LG that is the closest possible of the application and if possible with a different network path with the other LG's.

By doing this if you still have the same errors on all your LG's then i agree with Neil the issue is on application side. It could be the web server that does not accept anymore new connections. In that case when you reproduce those errors try to manually connect to the application to see if it works fine or not.

Yohann H.
Yohann H.

Yohann H.

Level
0
55 / 100
points

Thank you Neil, but the metrics of the test infrastructure seems to be ok and there are no errors in the Neoload logs.
Thanks also to you for your answer Nouredine, we will try to replay the tests without the antivirus to see if the problem could come from there. If not, we will set up a task force to study the network bricks with the people of the network at the customer.
Thank you guys for your advice.

Yohann H.
Yohann H.

Yohann H.

Level
0
55 / 100
points

For information, the problem is the same with the antivirus disabled on all injectors. Therefore, we will focus our analysis on the application environment. Thank you ;)

Yohann H.
Yohann H.

Yohann H.

Level
0
55 / 100
points

To keep the community informed, the problem came from the VIP F5 whose SSL license did not allow to go beyond the 500 transactions per second. Thanks for your help ;)