feat: storenode cycle #1223
feat: storenode cycle #1223
Conversation
richard-ramos
force-pushed
the
storenode-cycle
branch
from
d9361e5
to
3988fc6
Compare
| s.Lock() | ||
| defer s.Unlock() | ||
|
|
||
| for _, subs := range s.subscriptions { |
| m.host = h | ||
| m.failedRequests = make(map[peer.ID]uint) | ||
| m.peers = make(map[peer.ID]peerStatus) | ||
|
|
There was a problem hiding this comment.
Should we connect the store node first with some strategy before actually verify the status?
| } else { | ||
| m.peers[m.activeStorenode] = peerStatus{ | ||
| status: disconnected, | ||
| canConnectAfter: time.Now().Add(backoffDuration), |
There was a problem hiding this comment.
Another strategy is have a blacklist of unreachable store nodes, and periodically check if the node should be kept in the list.
| m.logger.Info("Automatically switching storenode") | ||
|
|
||
| if m.activeStorenode != "" { | ||
| m.disconnectActiveStorenode(graylistBackoff) |
There was a problem hiding this comment.
Wondering why this is needed and the intention of this graylistBackoff?
There was a problem hiding this comment.
Hm. Looking at the existing code in status-go, i believe it is so we dont connect again to this storenode for 3 minutes.
|
|
||
| m.activeStorenode = peerID | ||
|
|
||
| m.StorenodeChangedEmitter.Emit(m.activeStorenode) |
There was a problem hiding this comment.
I can't find the Subscriber of this emitter, do we still need it?
There was a problem hiding this comment.
waku/v2/api/history/cycle.go
Outdated
| return peer.status | ||
| } | ||
|
|
||
| func (m *StorenodeCycle) connect(peerID peer.ID) error { |
There was a problem hiding this comment.
How about rename it setActiveStoreNode?
| defer m.peersMutex.Unlock() | ||
| pInfo, ok := m.peers[id] | ||
| if !ok { | ||
| pInfo.status = disconnected |
There was a problem hiding this comment.
This will panic if no peer id found in the map.
There was a problem hiding this comment.
pInfo is not a pointer. Should be fine: https://go.dev/play/p/iCnMyu-hZpt
waku/v2/api/history/cycle.go
Outdated
| } | ||
|
|
||
| func (m *StorenodeCycle) verifyStorenodeStatus(ctx context.Context) { | ||
| ticker := time.NewTicker(1 * time.Second) |
There was a problem hiding this comment.
This can be a constant.
| m.logger.Info("trying performing history requests", zap.Uint("try", tries), zap.Stringer("peerID", peerID)) | ||
|
|
||
| // Peform request | ||
| err := fn() |
There was a problem hiding this comment.
I'm not familiar with this programming model, do see it in our codebase from time to time.
I'm thinking that will it be better if we use channels to communicate for the composing?
There was a problem hiding this comment.
Perhaps. I'd like to do this change in a separate PR if you agree!
richard-ramos
force-pushed
the
storenode-cycle
branch
from
b2d0226
to
b852a92
Compare
richard-ramos
force-pushed
the
storenode-cycle
branch
from
b852a92
to
f29ce5f
Compare
This PR contains the storenode cycle code from status-go.
I tried to simplify it a bit but most of the heavy refactoring should be done on a separate PR since this one is already complex enough.
@plopezlpz, @kaichaosun @igor-sirotin @chaitanyaprem:
I'm still working on the unit tests, but so far, while testing status-im/status-go#5857 with Desktop, it looks promising. Reviews are appreciated!