Memory usage on startup (old gen full)

Hello, everyone

I am facing a problem with my ES cluster, version 1.7.3
I have 10 nodes, each having 30g RAM.
Scatered around thos nodes are 2480 shards.

When I start the cluster, the heap usage skyrocket to 27-29g, with absolutely no cache being used. I'm not doing any indexing, nor do I search or aggregate on my data.
When I look at my nodes's stats, I can see that the old gen memory uses more than 28g.

If I start indexing, my cluster ends up collapsing due to increased time doing garbage collections that do not free any memory anyway.

I do not understand why I use so much memory.
Any insight would be greatly appreciated.

PS: please, be patient with my approximative english.

You should really upgrade to the latest 1.7 release.

How much data in your cluster?

Hello, and thanks for the answer

And yes .. I know I should upgrade, but frankly, If I had any say in the matter, I'd upgrade all the way to 2.3.

Anyway, I have news .. the memory usage plummeted to nice and good level after a while doing absolutely nothing on the cluster.

Since my nodes are running on virtual servers, and since one of the physical server crashed during some heavy indexation(making us lose a bunch of our nodes all at once), I suppose that a lot of memory was used on startup to get back to a stable state. (for the translog files or whatnot)

So the issue seems to have solved itself.
Anyway, thank again !