WEKAPP-281651

New QoS mount option to support controlling the number of client operations

Added a new mount option: 'qos_max_ops' that controls the maximum number of IO Operations per second.

This value is per mount point and is divided between the frontend processes servicing this mount point. Each frontend process has a minimum limitation of 100 ops/s.

If the specified value is lower than the minimum, it is adjusted to the minimum value.

WEKAPP-230789

Thin-provisioned filesystems in multiple organizations

Miscalculation of free space in non-root organizations without an organization quota and thin-provisioned filesystems might lead to failures in creating new (thick provisioned) filesystems in that organization.

Configuration: Multiple organizations and thin-provisioned filesystems

WEKAPP-191222

Improved error details when a CPU does not meet requirements.

The error message about mount operation failure due to a CPU that doesn't meet the requirements is improved with more information.

WEKAPP-261888

Cannot add a drive to a host that's on a different failure domain

The system does not prompt a message and does not prevent adding a drive to a host if the drive was previously configured with a different failure domain.

The solution includes adding a relevant message and preventing this action.

WEKAPP-26708

Packet corruptions cause assertion failures on the receiving process.

HW error-induced packet corruptions can trigger assertion failures on the receive side.

A new behavior prevents the asserts by banning the problematic server. With the Customer Success Team's assistance, you may manually override the restriction by using one of the following arguments:

* network_ban_disabled: Change to true or 1 to disable.

* network_ban_seconds: The time to ban the problematic server. The time is increased exponentially. That is, network_ban_seconds2^(num_occurrences-1).

network_ban_max_backoff_seconds: The maximum time to ban.

The ban is remembered in ephemeral (transitory) memory. A server restart of stop/start io causes all elements to forget the last ban state.

WEKAPP-285794

Fix baseline bug

Servers with laggy networking between them to the cluster can prevent the leader from taking some of the cluster management actions, this can be reflected in hanging stop-io, disk updates wouldn't progress, etc.

Impact: Cluster management action failures

WEKAPP-279828

False reservation of SSD capacity between thin provisioned filesystems

SSD capacity is falsely reserved for one thin provisioned filesystem while a different thin provisioned filesystem is filled.

WEKAPP-289452

Pausing a filesystem squash task might cause data inconsistency

Filesystem squash task is a task that is started automatically after downloading a snapshot. Pausing this task might cause data inconsistency.

WEKAPP-294491

Default quotas for deleted directories are not removed, they are kept in the default quota's table even though the directory has been deleted.

Default quotas for deleted directories are kept in the default quota's table even though the directory has been deleted.

WEKAPP-268022

IO operations hang when a directory quota reaches its exact value

When the directory quota is filled to its exact value, the QuotaHardLimitReached event is not issued, and IO operations to this directory may hang.

WEKAPP-269936

Default quota REST API returns a wrong inode_id

Default directory quota listing REST API returns the internal inode ID instead of the external one.

WEKAPP-232512

Resuming an object-store detach task while it is still pausing

Resuming a cluster task (object-store detach) while it is still in PAUSING state (visible in weka cluster tasks CLI) might cause the task to resume in an incorrect place. This might lead part of the detach task to be skipped, leading to data unavailability.

WEKAPP-277972

HCSF compute processes crash

HCSF compute processes might crash when reading invalid blocks instead of gracefully failing the IOs with an EIO error code (I/O error).

WEKAPP-276858

stat() after file rename
A rare race condition on a loaded client might cause a stat() call to return inaccurate information after a rename.

WEKAPP-23425

Negative directory entry cache is not invalidated on hardlinks

When using a non-zero dentry_max_age_negative mount option, directory entries that become positive due to an hardlink operation do not invalidate negative entries.

WEKAPP-254468

HCSF driver memory reclamation.

Some memory is not marked as cache and is not reclaimed back for other allocations at need.

WEKAPP-210697

Extend NFS interface name characters limit

Extended support for NFS network interface device names up to 14 characters long (previously 11 characters).

WEKAPP-260018

SMB service interuption

Sometimes, when one of the processes of the SMB service is restarted, it isn't automatically activated back. As a result, users get a Permission Denied message when trying to access shares or files through the SMB service.

WEKAPP-257495

Upgrade fails due to PKEY interface names parser issue

Parsing PKEY interface names has an issue preventing using HCSF alongside a PKEY configuration.

WEKAPP-270803

Incorrect drive states report.

The report displays incorrect drive states when running the weka cluster drive command.

WEKAPP-278557

HCSF processes hang after container restart

When working with E810 NIC, a container restart might cause the weka processes to hang in a SYNC state.

Configuration: E810 NIC

WEKAPP-296788

One-shot upgrade failure on clients

On one-shot upgrade mode to version 4.1.0, the upgrade will succeed on the backends, but may fail on the clients due to a timeout.

WEKAPP-277811

Clients roll back after upgrade

When using the upgrade-clients flag during the upgrade, the clients may roll back to the previous version a few minutes later.

WEKAPP-280878

A race condition may cause a failure during an upgrade

A race condition that causes clients' removal may lead to a failure in the upgrade process and not roll back to the source version.

WEKAPP-277811

Clients roll back after upgrade

When using the upgrade-clients flag during the upgrade, the clients may roll back to the previous version a few minutes later.

WEKAPP-280878

A race condition may cause a failure during an upgrade

A race condition that causes clients' removal may lead to a failure in the upgrade process and not roll back to the source version.

WEKAPP-278962

Upgrade rollback

Aborting an upgrade process (pressing Ctrl+C) does not roll back the upgrade process to the original version.

WEKAPP-296798

NDU failure

Client (stateless) mounting during NDU might fail the NDU process. The cluster is still up, and IOs continue.

WEKAPP-269693

Filesystem mount hangs after upgrade

A filesystem mount performed right after a weka local upgrade might hang on the client machine due to the HCSF agent failing to wait for frontend processes to finish starting up.

WEKAPP-295566

weka cluster nodes called from a client in a lower version

weka cluster nodes CLI command fails when called from clients that are on an older version (3.14.3) than the cluster version.

WEKAPP-283484

NFS permissions page does not load in the UI

When more than one client group is configured, the NFS permissions page does not load in the UI.

Hitachi Content Software for File User Guide

MK-HCSF000-01

October 2022

Hitachi Content Software for File Command Line Reference

MK-HCSF001-01

October 2022