[switch] Number of Stat objects created

[mcollina]

When running https://github.com/ipfs/benchmarks/blob/master/tests/local-transfer.js:

$ STAGE=local REMOTE=true FILESET=One64MBFile VERIFYOFF=true node local-transfer.js

That test uses only 2 peers.

I'm seeing around 44 Stats object being created, each one with their respective timer. Consider that at best (with libp2p/js-libp2p-switch#289), this means 44 events being fired every 2 seconds (the default). This is a lot of activity happening just to exists.

What is rationale behind keeping track of all those values? Are all these stats needed?

I've been looking at https://github.com/libp2p/js-libp2p-connection-manager/blob/0634c9116a3f708d3eb788ad7567022ec1f76195/src/index.js#L82, and the only Stats object that is needed is the global one https://github.com/libp2p/js-libp2p-switch/blob/f43084b93a1c048d68f7992357e3116029732ce2/src/stats/index.js#L40. Could the per-peer and per-transport stats be a) removed or b) put behind a flag?

[jacobheun]

Wow yeah, that's a significant amount. I've been digging into this a bit to figure out where else stats are being used. I know we want to make updates to the Connection Manager to have it make more informed decisions about closing stale/idle/low priority connections in the near future, so per peer stats will become more important there.

It looks like bitswap is getting it's data from the stats object ipfs/js-ipfs#1198. I'm not sure if the per peer or protocol is being used at the moment. Perhaps @hacdias and/or @pgte can provide some insight on where and which stats items are being used?

I think it's worth investigating if we need the per protocol statistics and if not, make those opt in. We should also look into more efficient ways of calculating the averages over time, especially if we need the extra stats on by default.

[pgte]

I don't know of any place we're using per-protocol statistics. IRC it was discussed at the time with @kumavis the need of per-transport stats only: #175 (comment)
Not sure that's changed..

[kumavis]

I ended up wanting really granular stats (transport/protocol usage per-peer) so I reused the Stats class but setup my own measurements

[wemeetagain]

We no longer have Stat objects, closing as stale