pass in kernel tbe id into rocksdb wrapper

pytorch / FBGEMM

FB (Facebook) + GEMM (General Matrix-Matrix Multiplication) - https://code.fb.com/ml-applications/fbgemm/

Other

1.17k stars 474 forks source link

pass in kernel tbe id into rocksdb wrapper #2930

Closed duduyi2013 closed 1 month ago

duduyi2013 commented 1 month ago

Summary: the reason we need this is we constantly see the port conflict error in rocksdb initialization. Before this diff we call getFreePort to ge an available port. For each ssd tbe we will create 32 rocksdb shards, so in total there are 256 ports needed per host. This works fine with 4 hosts until we are running 16 hosts training job as we need make sure all 16 hosts don't get into the corner cases where multiple db shard get assigned the same free port.

Differential Revision: D60635718

netlify[bot] commented 1 month ago

Deploy Preview for pytorch-fbgemm-docs ready!

Name	Link
Latest commit	639a2f71dc265a36ef7c83c2a7d3ec20a7a80671
Latest deploy log	https://app.netlify.com/sites/pytorch-fbgemm-docs/deploys/66b0128ebcb31a0008ce60a6
Deploy Preview	https://deploy-preview-2930--pytorch-fbgemm-docs.netlify.app
Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.