Unnecessary 'resume' of an input stream#525
Conversation
|
okay, i see... can i force to close the socket by my own? |
|
context: i have implemented an influxDB line-protokoll sever, to proxy this into another service. so i'm using the replaction API from indluxDB itself (https://docs.influxdata.com/influxdb/v2/write-data/replication/replicate-data/) to receive new data. the influxDB server is flooding me with sockets and data, and it seems that the connection is not instantly disconnected from influxDB server. so i have many open connection and hit the hard-limit of 1024 sockets in a short time. after some time, influxDB is closing the connection, but the connection seems never be reused. so may i can hard close this connection by my server. |
|
hmmm, i realy play around with this issue but i can not found a workaround with original code. maybe someone can put me in the right direction? i'm running all the time in the streams limit of 1024 ... the only thing what i can do is maybe put a load balancer in front of this and spawn some of my services? |

I have cleared up behavior here that I couldn't comprehend. A "resume" at this point leads to several streams that need to be managed again by the loop, which were previously removed by a
close-event. With a high number of client requests, the number of possible streams becomes too high, leading to a crash. The streams added back to the loop here eventually disappear "automagically," but this is not really comprehensible to me. Additionally, there is a significant increase in memory usage.