ReleaseNotes

The RozoFS development team is proud to announce version 1.3.2 of Rozo File System.

Changes in 1.3.2

---

Major Fixes:

• Fixed bug: avoid waiting indefinitely when reading a block is impossible (rozofsmount).
• Fixed bug when using rozo volume stat (rozo CLI).

Changes in 1.3.1

---

Major Fixes:

• Fixed bug: possible buffer overflow when opening a file (rozofsmount).
• Fixed bug when using rozo storage get command for one given storage (rozo CLI).

Minor Fixes:

• Remove undesirable output trace when using rozo volume expand command (rozo CLI).
• Fixed mistake in rozo man page for rozo volume remove command (rozo CLI).

Minor Enhancements:

• Increase the number of contexts for socketCtrl (storio).

Changes in 1.3.0

---

New Features:

• Add support for POSIX and BSD file lock (-o posixlock and -o bsdlock options).
• Core files handling:
  • Core files are generated on crash under /var/run/rozofs_core/APPLI/ for APPLI in rozofsmount, storcli, exportd, storaged and storio.
  • The maximum number of core files to keep on disk is optionally configurable per APPLI (default is 2). The oldest core files are kept.
    • -o nbcores option of rozofsmount plays for rozofsmount and for storcli,
    • nbcores parameter of exportd configuration file plays for exportd,
    • nbcores parameter of storaged configuration file plays for storaged and storio.
• Add Nagios plugins.

New Enhancements:

• New rozo manager CLI syntax and output: (rozo CLI)
  • rozo commands output are now in YAML format and are organized by topic.
• RozoFS constructor extended attribute:
  • A get on this attribute returns some RozoFS information on this file such as the export identifier (EID), the cluster (CID) and storages distribution (SIDs), file locks, and more.
  • A set on this attribute enables (when the file is created but empty) to modify the file distribution (SIDs).
• Synchronize read/write of applications on the same file and the same rozofsmount client. When an application writes any data to a file, this data is immediately accessible to any application on the same client. This is not true for applications on different clients that may reside on different servers.
• Inode number of a RozoFS file do not vary with mount points and mount time. Previously the inode numbering was local to the rozofsmount and chosen at lookup time.
• Adding various rozofsmount options:
  • -o rozofsnbstorcli: to set the number of storcli to run. Only the required number of storcli processes is started.
  • -o rozofmaxwritepending: to limit the number of parallel asynchronous writes pending per application. Usually most application write call are processed asynchronously. Once this number of write pending is reached, the application gets stuck on the write API until this last call is actually processed.
  • -o rozofsrotate: to configure rotation on non local servers for reading. This enables to balance the read requests on more servers.
  • -o rozofsminreadsize: to set the minimum data quantity to read from storages at a time.
  • other options for various timer configuration.
• rozofsmount/storcli uses the local server preferentially when reading a file that is distributed on the server where the client runs. This saves some network bandwidth.
• By-pass Mojette transform for empty blocks.
• Add cache for block mode.
• New software architecture for storage servers:
  • The storaged daemon starts only one storio process for IO services.
  • storaged and storio processes are now in non blocking mode.
  • Possibility to configure (via storage.conf) the number of thread(s) for IO requests.
  • Possibility to configure (via storage.conf) a list of ports on which IO services where provided.
  • storio automatically restarts after a crash.

Bug Fixes:

• Some error on the writing of the very first block of a file on disk may lead to block the file definitively.
• Add a check at rozofsmount startup to prevent double mounting when the exported file system contains no file.