Actions
Bug #12087
closed[arvados-ws] Memory leak
Story points:
-
Description
arvados-ws process caught using 40GB rss, 48GB virt, ~75% cpu.
debug.json:
{
"EventSource": {
"DBStats": {
"OpenConnections": 2
},
"EventsIn": 3479826,
"EventsOut": 54709791,
"Queue": 0,
"QueueDelay": 0.00097,
"QueueLimit": 4,
"Sinks": 44,
"SinksBlocked": 0
},
"HTTP": {
"ReqsReceived": 30041,
"ReqsActive": 45
},
"Outgoing": {
"QueueCount": 191,
"QueueMin": 0,
"QueueMax": 64,
"QueueTotal": 9204,
"QueueDelayMin": 0.000161,
"QueueDelayMax": 5.133126
}
}
An arbitrary log entry:
{
"RequestID": "bevbt12eapbb",
"elapsed": 872.1006494,
"level": "info",
"msg": "disconnect",
"stats": {
"QueueDelayNs": 10791625100,
"WriteDelayNs": 49772000,
"EventBytes": 904645,
"EventCount": 1519
},
"time": "2017-08-08T15:51:38.274834635Z"
}
A quick look at strace just shows an awful lot of clock_gettime() between useful-looking reads and writes. Perhaps some goroutines are getting stuck in busy loops?
Actions