On 16.07.2018 16:35, David Cleary wrote: > 2018-07-16 15:55 GMT+03:00 David Cleary <da...@progress.com>: >>> We have a customer who is experiencing a random, 21 second pause when using >>> out Tomcat > based application server. We believe this may be during a TCP connect and > timeout. Logging > indicates the pause happens before the request makes it to our back end.
> Logging where then ? Sorry for any formatting issues. I have a digest subscription which doesn't lend well to interactivity. Clients are running on Windows machines. Server is running on AWS and Linux. There is a cloud firewall in between (pfSense). Do not have the details if they are running the cloud version available on AWS. Client logging shows this: [18/05/16@12:12:48.822+1000] P-006760 T-002372 1 4GL REV Trying Connection [18/05/16@12:13:09.925+1000] P-006760 T-002372 1 4GL REV Connect Complete 21102 [18/05/16@12:13:09.925+1000] P-006760 T-002372 1 4GL REV WARNING: LONG CONNECTION [18/05/16@12:13:09.925+1000] P-006760 T-002372 1 4GL REV A4DC513EA548E24508E1E90AA9EA61DD9386DDB475AD.clintons connected 21102 Localhost access log shows this localhost_access_log.2018-05-16.txt:10.255.11.250 - - [16/May/2018:12:13:16 +1000] "POST /apsv?CONNHDL=A4DC513EA548E24508E1E90AA9EA61DD9386DDB475AD.clintons HTT P/1.1" 200 253 1 And our back end agent log shows this: clintons.agent.log:[18/05/16@12:13:16.294+1000] P-019364 T-2819262208 2 AS-19 AS Application Server connected with connection id: A4DC513EA548E24508E1E90AA9EA61 DD9386DDB475AD.clintons. (8358) clintons.agent.log:[18/05/16@12:13:16.299+1000] P-019364 T-3688318720 2 AS-19 AS Application Server disconnected with connection id: A4DC513EA548E24508E1E90AA9E A61DD9386DDB475AD.clintons. (8359) Customer had some weird reconnection logic that was part of their application. After removing the code so the logical connection would be kept open, we saw this pause happen on a standard request. I do not know how long this logical connection was idle before running. I also do not know if Tomcat closed the underlying socket either due to resources or a keep-alive timeout. I was hoping logging could tell me when Tomcat binds to an incoming socket and releases it. I was hoping to show in the above example, as far as Tomcat is concerned, the 21 second delay happened outside of the server. Scouring the source code and trying some experimentation, it does not appear there is logging available at the socket level. It mostly happens > when we create an initial logical connection, but we have also seen it > elsewhere where we > believe the TCP Keep alive was expired and a new socket had to be > established. However, I > do not know this and am hoping there is some logging I can turn on in the NIO > connector to > collect more data. I tried turning on logging in the Endpoint class, but that > did not provide > anything useful. > If the connection request does not even reach the Tomcat back-end, that is > also unlikely > to provide much information. (Not being facetious here, just stating a fact). > Can you do a "netstat" command on your Tomcat server when this happens ? > If yes, maybe some part of the output would provide some information from the > TCP level > (such as a high number of connections, to the Tomcat NIO port, in some > specific TCP state e.g.) Customer did some probing with Wireshark and said they were seeing a TCP retransmission (sorry, I do not have many more details). In investigating this, we discovered this info on TCP timeouts: " There's probably a million reasons why the client may never receive a SYN-ACK. The one I've seen more often is packet loss, which in turn can have lots of reasons, for example a malfunctioning or misconfigured network switch. However, you can immediately spot if your timeout/hang problems are caused by TCP retransmission because they happen to cause response times that are unusually frequently distributed around 3, 9 and 21 seconds (and on, of course). In fact, the TCP retransmission timeout starts at 3 seconds, but if the client tries to resend after a timeout and still receives no answer, it doubles the wait to 6 s, so the total response time will be 9 seconds, assuming that the client now finally receives the SYN-ACK. Otherwise, 3 + 6 + 12 = 21, then 3 + 6 + 12 + 24 = 45 s and so on and so forth." This is why we are focusing on the TCP layer. > There is a NAT firewall between the client and server, so I'm looking for > some TCP level logging that could point me in the proper direction. > >> Tomcat version = ? > > Sorry. Tomcat 8.5.27. > > And on which kind of O.S. is this happening ? > Also maybe another question : is this happening on a Tomcat server which is > dedicated to > that particular customer ? or is the Tomcat server shared between different > customers, and > only that particular customer experiences these delays ? We sell an application server that customers create their own applications on. This particular customer has many customers themselves. The customer's application do not exhibit this running on our older, non-Tomcat based AppServer. Since this happens randomly, only a couple of times a day, it is difficult to diagnose. Since the customer doesn't see this issue, running the same exact client, on our older AppServer, they believe it is the new one. However, the older one isn't HTTP-based, and they had a bunch of hacks related to connections where they would recycle a connection after 5 minutes of inactivity, or after a 30 minute lifespan. This says to me the issue clearly isn't our appserver, but I can't prove it at this point. The firewall is my likely culprit, but without logging at the Tomcat endpoint, I can't definitively say where the pause is. --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
