[ https://issues.apache.org/jira/browse/FLINK-32878?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17792115#comment-17792115 ]
Martijn Visser commented on FLINK-32878: ---------------------------------------- I reverted this change via c7da98f23c3f86de3a3b12355fa7a1289200f93d because CI indicated that it had failed and I couldn't compile it locally. However, CI was failing because of an unrelated error (download issue on NPM) and my local version failed because of my mistake. So re-enabled it via 87d7b4abd0a1ec92433603c83401cc8ad00fd500 > Add entropy to gcs path for better scalability > ---------------------------------------------- > > Key: FLINK-32878 > URL: https://issues.apache.org/jira/browse/FLINK-32878 > Project: Flink > Issue Type: Improvement > Components: Connectors / FileSystem > Affects Versions: 1.15.4, 1.16.2, 1.17.1 > Reporter: Jayadeep Jayaraman > Assignee: Cheena > Priority: Major > Labels: pull-request-available > Fix For: 1.19.0 > > > Currently GCS is used as a backend for both checkpointing and sink. In both > these cases the file names are sequential which causes hotspotting in GCS and > results in HTTP 5XX status code. > > As per [GCS best > practices]([https://cloud.google.com/storage/docs/request-rate)] it is > advisable to spread out the object creation with random names close to the > beginning of the file name. > > There is similar work done already for S3 as part of FLINK-9061 -- This message was sent by Atlassian Jira (v8.20.10#820010)