You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
split S3 files into smaller files to send large union file (facebookresearch#77)
Summary:
Pull Request resolved: facebookresearch#77
# Context
We found that AWS-SDK S3 API would fail when we try to write more than 5GB of data. It is a blocking us to do capacity testing for a larger FARGATE container.
In this diff, as mentioned in [the post](https://fb.workplace.com/groups/pidmatchingxfn/posts/493743615908631), we are splitting union file based on number of rows.
# Description
We have made following changes.
- Added new arg `s3api_max_rows` in the private-id-multi-key-client and private-id-multi-key-server binaries. We will use this to split a file for S3 upload.
- Added an optional arg `num_split` in save_id_map() and writer_helper(). When `num_split` is specified, it would use the arg `path` as its prefix and save files in `{path}_0`, `{path}_1`, etc.
- In rpc_server.rs and client.rs, calculates the num_split based on s3api_max_rows, and passes the num_split arg for S3 only. Then, for each split file, it calls copy_from_local().
Differential Revision: D39219674
fbshipit-source-id: 82dc1788b0d4db5cf9c3de07178b52a8cc11633c
0 commit comments