Split object_space_dump into smaller messages

[emilsoman]

Closes #52

Merge after #76 is merged.

Stuff that changed :

1. Added snapshot_no to obj space dump events (so these can be grouped even if they arrive in different messages)
2. Accumulated events will be sent prematurely(before the 1 second gap) if we have more than a fixed number of events in the queue (set to 400 for now after some experimentation)
3. Added a limit of 20 object dumps to be grouped into one event. So we'll at a time receive a max of 20*400 object dumps in one zmq message.

[gnufied]

@emilsoman can you document Message format after this change? Potentially in a new WIKI page so as not to remove existing one.

[emilsoman]

@gnufied , the only change is a new key that's added to the payload of each object space dump message called "snapshot_no" , I've added this to the wiki already

[gnufied]

@emilsoman can you update this PR so as it is mergeable?

[emilsoman]

@gnufied , done.

[gnufied]

20 is very small number emil and I think for larger projects we will see drop in messages, because queue will be too full.

[gnufied]

I think this number has to be way higher, but it has to be calibrated. But something like 1000 should not be too bad of a limit IMO.

[gnufied]

@emilsoman can we update this PR with latest master?!

[gnufied]

Okay, so this looks mostly good. One problem though is - this specifically implements splitting of heap dumps which is fine.

But I think, we will require similar splitting of larger cpu profiling dumps, so it makes sense to make it bit generic.

I think it makes sense to add a new field called correlation-id to all events, for other events the value can be anything (Sometime like Time.now.to_f.to_s), but for events that can be split, they will have same correlation-id . So if a snapshot event is being split across several messages, all those event messages will have same correlation-id.

What I am really proposing is simply renaming snapshot_no to correlation_id, but I think it makes entire thing more simpler.

[gnufied]

On seconds thoughts, adding correlation-id to all events may not be necessary, but just to events which can be split is fine!

So really, my only request is to just rename the field itself. :-)