At the moment, every time there's a request for the predictor, there's a hit on the model store to check for updated models.
In a high traffic scenario this can lead to a degradation of the predictor's responsiveness.
In this PR a configurable time-between-checks in milliseconds is introduced via the env var MODEL_STORE_CHECK_RATE (e.g. a value of 60000 will check for updated models every minute).
The previous behaviour (a check with every request) can be simulated by setting the interval to 0.
ruivieira
commented
6 years ago
At the moment, every time there's a request for the predictor, there's a hit on the model store to check for updated models.
In a high traffic scenario this can lead to a degradation of the predictor's responsiveness.
In this PR a configurable time-between-checks in milliseconds is introduced via the env var
MODEL_STORE_CHECK_RATE(e.g. a value of60000will check for updated models every minute).The previous behaviour (a check with every request) can be simulated by setting the interval to
0.@elmiko @zmhassan @sophwats ptal.