Open MrOrz opened 4 years ago
Regarding the "Very long string" error emitted by elasticsearch and triggers OOM, I highly suspect that it is because rumors-api is trying to access html
field in urls
index, and that triggers such error.
However we can never be sure and further investigation is needed (such as analyzing field size of each field in elasticsearch) to develop a step-to-reproduce, and then a solution (like blacklisting certain fields for searchResultLoader
and docLoader
) can be applied.
On 2/28 5:03 we had a downtime:
On rumors-api stdout & stderr, it seems that the system has gone through several OOM error and restarts.
This log is also reoccurred hundreds of times.
See the full logs here: https://gist.github.com/MrOrz/63019de0f65fd320964770dd64bebbe0
rumors-api:0
&rumors-api:1
encounters JS heap OOM and restarts sparainglyrumors-api:0
&rumors-api:1
encounters JS heap OOM and restarts alternatively. Each restart is within 4 seconds and there are ~ 30s gap between restarts of different processes, thus the service should be still available.