> On 20 Aug 2018, at 16:40, Aleksandar Lazic <al-hapr...@none.at> wrote: > > Hi. > > Am 20.08.2018 um 15:28 schrieb Ing. Andrea Vettori: >> Hello, >> >> we’re using haproxy since version 1.7 for our websites. Backends are mostly >> Apache Tomcat. We have a centralised log server for application logs. >> A couple weeks ago we upgraded to 1.8.13 and enabled h2. Since then, in our >> logging system we started to see a number of 500 responses. What’s strange >> is that users didn't notice any issue and “near” a 500 response there’s >> almost always a 200 response for the same URL. Looking further into the >> problem we found that 500s are caused by what java reports as IOException >> for ‘Broken Pipe’. >> We find such errors for a number of client user-agents so it seems not >> “client” related or at least if it’s a client problem it’s widespread. >> >> Is this a problem that can be related to how haproxy works with the h2 >> protocol ? Is it possibile that for some reason it’s haproxy that is calling >> the same url two times and one of the connection is then interrupted ? > > Maybe it have something to do with this > > https://cbonte.github.io/haproxy-dconv/1.8/configuration.html#4.2-option%20http-ignore-probes > > Does the behavior changes when you set this option?
Hello and thanks for your help. I didn’t tried that option yet and will do tomorrow. However I’m not sure to have clearly described the problem we’re noticing. The multiple calls are seen in our centralised log system, not in haproxy logs. I’ll try to look at haproxy logs configuration to log more than what we’re doing today (only warnings at the moment). Also the “double calls” appear to be random and in most cases they are in the middle of browsing sessions. And they’re separated in time by milliseconds, the 500 first, the 200 second, so they definitely appear to be done by a computer program, not by a user clicking two times on a link. Thanks — Ing. Andrea Vettori Responsabile Sistemi Informativi
