Fix autoBreadcrumb http instrumentation memory leak

[LewisJEllis]

PR #276/version 1.1.3 introduced a memory leak in our http instrumentation as reported in #295. The core issue was that we were monkeypatching http.ClientRequest.prototype.emit during every ClientRequest instantiation, instead of just during initial instrumentation, which meant that for each outbound http request made, we kept a bunch of stuff around in a closure.

Users on raven-node 1.1.3 or 1.1.4 with long-running processes that made outbound http requests found their memory usage growing steadily over time.

This PR refactors the ClientRequest instrumentation to only monkeypatch http.ClientRequest.prototype.emit once during initial Raven setup when we also monkeypatch the constructor. The monkeypatched emit method previously found the request url by closure reference, but now the constructor attaches a __ravenBreadcrumbUrl property to the request object and emit grabs that instead.

I confirmed the issue and fix using memwatch-next (plus express, some outbound requesting, and some apachebenching) and have ideas for how to use that to automate some "did we intro a memory leak" sanity testing, but that can come later. That effort will have to test multiple different possible causes; the manual memory load test in https://github.com/getsentry/raven-node/blob/master/test/manual/context-memory.js would not have caught this because it only exercises contexts, not each of the different autoBreadcrumb instrumentation options.

This should fix #295. As soon as this merges I'll publish a new version.

/cc @benvinegar @MaxBittker @maxcountryman

[maxcountryman]

Nice catch! 👍

[benvinegar]

Let's cut a new patch release.

[idris]

Would love to see automated tests around memory leaks, as you mentioned in the description.

This is the second time Raven introduced a memory leak into my app, with the first being #231

[maxcountryman]

I would like to second @idris comment with a big plus one: Raven took down my production nodes due to this problem and it would be nice to see provisions against this happening going forward.

[oliviertassinari]

@LewisJEllis Thanks for fixing it 🆗

[emmenko]

Thanks for the fix, really a nasty one!!! 🎉

FYI: this is how it looked in our grafana dashboards: