How should I go about debugging slow response times for sockjs requests? By slow, I mean 20-30 seconds. We use AWS ELB for load balancing and occassionally see crazy spikes in response times (between ELB and our server) for sockjs requests. We use Kadira and can see there is no correlation between instance CPU usage, memory or number of sessions and these slow requests. We’ve also checked the AWS systems logs and see no problem with network or disk performance during these windows of slow response times.
One thing I noticed when looking at the access logs is the client IP address for the slow requests is the same for each “batch”, e.g., it would seem that sometimes a single client gets “slowed down” and those requests remain slow for quite some time.
Could a meteor package (or me, I suppose) inadvertently be attaching some “hook” to sockjs requests on the server, that slows things down for specific users/clients?