Heroku deployment response time alert

[victorlin]

Over the past few days we have been getting several emails (e.g.) with the following message:

The 95th percentile response time for nextstrain-server has exceeded your threshold setting of 1000ms.

Historical response time data is available in Application Metrics. Additional guidance on correcting slow requests and request timeouts is available in Dev Center.

They have all self-resolved within minutes, but still out of the ordinary.

Here is a screenshot of metrics over the past week. I've turned off max response time to make the percentiles more readable.

Tasks

1. Investigate difference in behavior
2. If this is expected, increase the alert threshold.
   Otherwise, take action to reduce response times.