Expose the same metrics as Kubo

[lidel]

Kubo Gateway exposes a bunch of metrics on RPC port: http://127.0.0.1:5001/debug/metrics/prometheus

bifrost-gateway should do the same

Details

We already have the some code extracted from Kubo:

https://github.com/ipfs/go-libipfs/blob/302b2799386dea7afb72ba0b4c32a5c427215d06/gateway/handler.go#L220-L262

We need to also move things from MetricsCollectionOption:

https://github.com/ipfs/kubo/blob/14649aa8ba8d7612ce9e35bba776fe7e7498b343/core/corehttp/metrics.go#L79

While at it, we should fill the gaps mentioned in ipfs/boxo#154

[lidel]

@gmasgras is that all you need for drop-in replacement? Is there a place we can find all necessary metrics, to double-check we did not miss anything?

[willscott]

for caboose preference for using Prometheus metrics directly versus opencensus?

[gmasgras]

Here is a list of metrics we're currently using for our dashboards and alerts

ipfs_bitswap_active_block_tasks
ipfs_bitswap_active_tasks
ipfs_bitswap_pending_block_tasks
ipfs_bitswap_pending_tasks
ipfs_bitswap_recv_all_blocks_bytes_count
ipfs_bitswap_recv_dup_blocks_bytes_count
ipfs_bitswap_sent_all_blocks_bytes_sum
ipfs_bitswap_wantlist_total
ipfs_http_response_size_bytes_sum
ipfs_http_unixfs_get_latency_seconds_sum
ipfs_p2p_peers_total
libp2p_rcmgr_conns_blocked_total
libp2p_rcmgr_memory_allocations_blocked_total
libp2p_rcmgr_peer_blocked_total
libp2p_rcmgr_protocols_for_peer_blocked_total
libp2p_rcmgr_services_blocked_total
libp2p_rcmgr_streams_blocked_total

[willscott]

Of these, only

ipfs_http_response_size_bytes_sum
ipfs_http_unixfs_get_latency_seconds_sum

are relevant

[lidel]

Thanks!

Hm.. it looks like we don't use existing metrics for directory listings, nor blocks, cars, and miss metrics for newly added types like JSON/CBOR/IPNS.

We'll clean this up in ipfs/boxo#154 and then document all metrics exposed by bifrost-gateway in its readme.

[lidel]

@hacdias #10 exposes ipfs_http_unixfs_get_latency_seconds_sum but ipfs_http_response_size_bytes_sum is missing, because it was not part of gateway code. Mind tracking down how it is exposed in Kubo and adding it to bifrost-gateway? We need parity here. 🙏

[lidel]

Last one landed in #18