Skip to content

3.23
Compare
Choose a tag to compare
@chrislusf chrislusf released this 22 Aug 02:34
· 2053 commits to master since this release
3.23
c4e862e

File content uploading now is more robust. There are two steps, assigning a file id from master and uploading to volume server. Previously the two steps are usually retried independently. However, it is possible that after step 1, the volume server will change the volume to read only, especially during volume operations, e.g., rebalance, erasure coding, etc. In this release, the two steps are retried together. This should help for filer.copy, mount, webdav, etc.

Another fix is Erasure Coding for volume server, when a volume server has multiple directories and a shared index directory. This combination caused some left over files, which caused other volume manage problems.

  • Filer
    • Retry read requests if not found in one replica
    • Fix if-modified-since check doesn't fire on "exact" #3467
    • Fix file uploading for /etc #3476
  • S3
    • Fix regression on HEAD directory operation #3457
  • filer.backup
    • backup small files if the file is saved in filer (saveToFilerLimit > 0) #3468
  • Mount
    • Implement copy_file_range syscall #3475
    • Fix buffer leaks in file gaps. #3472
    • Fix hanging read goroutines in chunk cache #3473
    • Avoid race conditions with current filer address #3474
  • Volume
    • Fix metrics reporting on volume file count.
    • Fix EC encoding/decoding problem when multiple dir locations and shared index directory are used together.
  • Minor
    • filer.sync offset initialization #3450 #3451
    • S3 fix metrics reporting for deleted collections #3453
Assets 34