Check for new added nodes #512
Conversation
AbdelrahmanElawady
requested review from
muhamadazmy,
MarioBassem,
ashraffouda,
rawdaGastan,
AlaaElattar,
xmonader and
sameh-farouk
as code owners
| @@ -29,6 +30,7 @@ type NodeGPUIndexer struct { | |||
| batchSize int | |||
| nodesGPUResultsChan chan []types.NodeGPU | |||
| nodesGPUBatchesChan chan []types.NodeGPU | |||
| newNodeTwinIDChan chan []uint32 | |||
There was a problem hiding this comment.
Why is that a channel of int32 slice? this should keep receiving the latest twinID inserted on the nodes table, no?
There was a problem hiding this comment.
ah forget it checks the nodes between the last id you checked and the last id on the table, could be more than one indeed
| @@ -934,3 +934,9 @@ func (p *PostgresDatabase) UpsertNodesGPU(ctx context.Context, nodesGPU []types. | |||
| } | |||
| return nil | |||
| } | |||
|
|
|||
| func (p *PostgresDatabase) GetLastNodeTwinID(ctx context.Context) (uint32, error) { | |||
There was a problem hiding this comment.
Maybe we should change it to be GetGreaterThanTwinID and use the twinID we checked last? this should get us the list of ids that we can directly send?
There was a problem hiding this comment.
Initially, we thought we only need the latest ID and all new nodes will be from last checked to the latest ID however this is not right as twin IDs are shared with other entities so node twin IDs don't have to be incremented one by one. So, this approach is probably better.
| ticker := time.NewTicker(newNodesCheckInterval) | ||
| latestCheckedID, err := n.db.GetLastNodeTwinID(ctx) | ||
| if err != nil { | ||
| log.Error().Err(err).Msg("failed to get last node twin id") |
There was a problem hiding this comment.
This means the proxy is having a seriou issue where there're no nodes in the table? or could that while the pod is starting for the first time?Is that the reason why we don't panic here?
There was a problem hiding this comment.
I think there might be non-fatal errors here that would not stop the operation of the indexer, that's why we don't panic. What I can think of is maybe some network issue or db connection error on start. I think we don't need to panic here.
| n.newNodeTwinIDChan <- nodeTwinIDs | ||
| latestCheckedID = int64(nodeTwinIDs[0]) | ||
| case <-ctx.Done(): | ||
| break |
|
Thank you 💯 |
Description
Check for new added nodes periodically and update GPU data when new nodes are added.
Related Issues
Checklist