Unless specifically noted otherwise, this history of problems fixed for IBM Spectrum Scale 4.2.x applies for all supported platforms.
Changes to this release of the IBM Spectrum Scale licensed program and the IBM Spectrum Scale library include the following:
- Enhanced operating system support with spectrumscale installation toolkit
The spectrumscale installation toolkit now also supports the following operating systems: Red Hat Enterprise Linux 6.8 and SLES 12 on the Intel x86_64 architecture Red Hat Enterprise Linux 6.8 on the PPC64 architecture Red Hat Enterprise Linux 7.1 and 7.2 on the PPC64LE architecture
- Enhanced Support for accessing data over SMB without setting the READ_ACL ("Read Permissions") bit on every file and directory
Files and folders that have not been accessible previously due to missing READ_ACL (gpfs) / "Read Permissions" (win) rights can now be accessed as long as the ACL is not accessed (given that the other access rights like READ grant access).
- Extended status information
The subroutines gpfs_fstat_x() and gpfs_stat_x() provide extended status information. The subroutines directly return a gpfs_iattr64_t structure that contains many additional fields that are available only through this call, such as file creation time and file generation time.
- GPFS log time stamp with time zone information
The default time stamp format for the GPFS log now includes time zone information and is similar to the ISO 8601 time stamp format. With the new format, you can convert times unambiguously to absolute times and you can sort and merge entries more easily. You can switch between the new log time stamp format and the previous format with the mmfsLogTimeStampISO8601 attribute of the mmchconfig command. You can also specify the log time stamp format for the entire cluster or for individual nodes. If you are migrating to v4.2.2, you can avoid automatically switching to the new time stamp format by specifying the mmfsLogTimeStampISO8601 parameter when you run the command mmchconfig release=LATEST.
- Transparent cloud tiering
The following changes are made: Support for cloud data sharing. It allows data to be moved across disparate geographical locations and/or heterogeneous application platforms. Support for multiple node classes. You can enable and manage independent groups of Transparent cloud tiering nodes in different node classes for use with different network configurations per node class. Support for multiple file systems where one node class points to one file system. Support for the mmcloudmanifest tool, which you can use to parse the manifest file that contains the list of files exported to the cloud. Support for displaying the cloud service version installed on each node in a node class. Support for IBM Cloud Object Storage on IBM SoftLayer. Support for locally displaying the thumbnail of files without recalling them from the cloud storage tier. Support for configuring the number of days for which the deleted files are to be retained on the cloud.
- IBM Spectrum Scale GUI changes
The following main changes are made in the IBM Spectrum Scale management GUI: Added new Home page. The Home provides an overall summary of the IBM Spectrum Scale system configuration and health status of its components and services that are hosted on it. Added new Files > Transparent Cloud Tiering page. The Transparent Cloud Tiering page provides both summarized and attribute-wise details of the Transparent Cloud Tiering service, which is integrated with the IBM Spectrum Scale system. Added new Storage > NSDs page. The NSDs page provides an easy way to monitor the performance, health status, and configuration aspects of the all network shared disks (NSD) that are available in the IBM Spectrum Scale cluster. File system detailed view is added in the Files > File Systems page. The detailed performance view helps to drill-down to various performance aspects. To access the detailed view, select the file system for which you need to view the performance details in the File Systems page and select View Details. Support for external keystone server object user authentication is added. You can now configure either an internal or external keystone server for authenticating the object users. The IBM Spectrum Scale system health component, mmhealth, replaces internal GUI health monitoring system. All the GUI panels and components that display the system health status are modified to display the health status reported by mmhealth. Introduced component-wise email notifications. The system administrator can now configure the email notifications in the Settings > Event Notifications page to send email notifications to the recipients, if events are reported in the following functional areas: Authentication Block and iSCSI services CES network Transparent Cloud Tiering NSD File system GPFS GUI Hadoop connector Keystone Network NFS Object Performance monitoring SMB Object authentication Node CES The Simple Network Management Protocol (SNMP) Management Information Base (MIB) is modified.
- Improved performance monitoring using Grafana:
You can use the Grafana tool to analyze and display performance data. Grafana is an open source tool that uses a performance monitoring bridge to set up and populate graphs, which can be easily viewed and analyzed.
- Logging file system activity using Varonis:
File system activities can now be logged using the Varonis DatAdvantage software.
- Mixed operating systems support with spectrumscale installation toolkit
You can now use the spectrumscale installation toolkit to install GPFS and deploy protocols in a cluster that contains nodes that are running on different operating systems.
- Network checking: mmnetverify command
With the mmnetverify command, you can verify the network configuration and operation of a group of nodes before you organize them into an IBM Spectrum Scale cluster. You can also run the command after you create a cluster to analyze network problems. Tests include address checks, ping tests, remote shell and file copy tests, time-date checks, TCP connection checks, message size tests, bandwidth tests, and flooding tests. The command prints full information and error logs about all the nodes tested.
- New sensors added to the list of performance metrics:
The following three new sensors have been added to the list of performance metrics GPFSFileset GPFSPool GPFSWaiter
- Quality of Service for I/O operations (QoS)
The mmchqos command has added capabilities. You can assign IOPS to individual nodes, to a node class, to a list of nodes in a text file, or to a remote cluster. You can also preserve and reuse your IOPS assignments by configuring them in a stanza file.
- RDMA CM (Connection Manager) and RoCE improvements
You can now create multiple IP network interfaces per RDMA device and per VLAN. With RDMA over Converged Ethernet (RoCE), you can now create virtual IP network interfaces (such as eth0:1) and IP network interfaces with multiple IP addresses.
- REST-style API for performing IBM Spectrum Scale tasks
The IBM Spectrum Scale REST API is an HTTP programming API for performing command-level IBM Spectrum Scale tasks. With the REST API, you can automate storage management operations and integrate IBM Spectrum Scale capabilities into your applications. The API can be installed on a single cluster node and requires an Apache server to be installed on the same node. It runs on HTTPS and uses JSON syntax to frame data inside HTTP requests and responses. In this release the API supports only high-prioriy operations, including operations on filesets, snapshots, and quotas and retrieving information about CES objects, file systems, and nodes.
- Documented commands, structures, and subroutines
New commands The following commands are new: mmnetverify mmrest New structures There are no new structures. New subroutines The following subroutines are new: gpfs_fstat_x gpfs_stat_x Changed commands The following commands were changed: mmchconfig mmchqos Changed structures The following structures were changed: gpfs_iattr64_t Changed subroutines There are no changed subroutines. Deleted commands There are no deleted commands. Deleted structures There are no deleted structures. Deleted subroutines There are no deleted subroutines.
- Object storage improvements
Added support for the object heatmap data tiering policy. Added support for secure communication between the proxy server and the other backend servers. Added support for enabling object access on existing filesets.
Problems fixed in IBM Spectrum Scale 4.2.1.2 [November 3, 2016]
- Fix problem with choosing only fully cached files for Cache Eviction.
- Fix a problem in which mmfsck ends prematurely with exit status 666:9:2.
- Fix a mmcrsnapshot command hang. This would only happen if a snapshot exists before and after the inode space expanded on a problem file system or fileset.
- Fix a deadlock that can occur because of a race condition between dmapi calls with file handles and a file deletion.
- Fix a problem in which the node freezes and it is required to be restarted. This can occur when vfs hits it's open file limit found in /proc/sys/fs/file-max.
- Fix an ioctl error that can occur when running the mmgetstate -a command.
- Improve consistency check for CES-only related IP movements.
- Fix a problem in which different up-calls are dropped.
- Fix a faulty Events Exporter assertion. The faulty assertion could cause the mmfsd daemon to fail in rare instances while generating an internal dump or responding to certain "mmfsadm saferdump" commands.
- Provide a new config option afmSyncAllXattrs to synchronize all xattrs.
- Fix a "cryptographic library could not be initialized" error that can occur when changing a cipherList in a P8LE environment.
- This fix improves GNR's error handling of pdisk faults that can occur after the fault-tolerance of the system is exceeded.
- Fix the 'pStepReceiveP->phase == phase' assert that can occur when fsck is run in a tight loop.
- When there are tons of users and groups intertwined with a lot of filesets with --perfileset-quota enabled this fix can improve the speed of mmcheckquota greatly.
- Fix a problem in which an offline fsck overflowed a duplicate fragment buffer.
- Fix a 'opP != __null' assert hit by offline fsck when handling a corrupt compressed files.
- Fix a mmap read issue that can occur when getting stale data from the kernel page cache.
- Fix a problem in which a NFS client gets an IO error. This can occur in a Ganesha environment.
- Fix a problem in which fcntl lock waiters are not resumed when a linux holder downgrades.
- Fix a problem in which 3524 storage enclosures are excluded when syncing storage enclosures. This fix applies to GSS/ESS customers.
- Fix a problem in which mmrepquota shows the fileset id but does not show the name of the added fileset quota.
- Fix a problem which is causing a disconnect between the AFM DR pairs. This can occur using the mmcesdr command after initial configuration has been completed on both clusters.
- Fix a "[Errno 2] No such file or directory" error that shows up in the /var/adm/ras/mmsysmonitor.log when the sqlite3 and the tar executables do not exist in /usr/bin/tar and /usr/bin/sqlite3. This can cause the FILESYSTEM and DISK monitors of Sysmonitor to get into an UNKNOWN state.
- Fix the exp(!ebP->replyDone) assert that can occur when DMAPI is being used.
- Fix a problem in the AFM environment where replication would stop because of an error that might occur while replaying Rename operations. This can cause AFM filesets to go into a recovery and/or resync state.
- Fix a problem in which IBM Spectrum Scale Object Protocols functionality is affected by security vulnerabilities in the TLS and SSL protocols (CVE-2015-2808 and CVE-2014-3566). This fix is applicable only if Object is enabled and SSL for keystone is configured.
- Fix a slow down in offline fsck repair. This occurs when inodes in a fileset are not deleted when an afm fileset is converted to normal.
- Fix a file system hang that occurs during mounting when migrating from an old version of GSS/ESS. This can occur when the logtip vdisk was created with a blocksize of 256k.
- Fix a problem in a cluster configured with CCR, where a node which is rebooted may be unable to rejoin the cluster. After the reboot, files /var/mmfs/ccr/ccr.paxos.1 and /var/mmfs/ccr/ccr.paxos.2 may have a length of 0. The problem tends to happen in VMs. Error message "Not enough CCR quorum nodes available" (err 809) may be produced for most commands.
- Fix offline fsck reporting false positive directory entry problems.
- If some physical disks in GPFS Native RAID (GNR) suffer a large number timeouts (very slow IOs), they should be declared "slow", drained and scheduled for replacement. This defect may have prevented them from being declared slow, meaning they will continue to lower performance and risk outages.
- Fix a deadlock that can occur when a pagepool is too small in a RDMA enabled environment.
- Fix a problem in which a write fails on a file which does not have any remote Pcache EA's set. This will cause the fileset to need to be recovered later.
- Fix a problem on LROC devices greater that 4TB in which the display of mmdiag lroc shows an incorrect Max Capacity value.
- Fix a "FSErrBadDittoAddr" file system struct error that can occur on a clone child file.
- Fix a problem in which extra snapshots are created. This can occur if an error 5 occurs while creating a snapshot at home.
- Fix a gateway node crash that can occur if it accesses garbage content being passed in from gpfsReadAfmDRLastRPOSnapName.
- Fix an error 124 which can occur if the control interface is not set up properly during DR setup.
- Fix a setTime assert. This can occur if mmap is being heavily used.
- Fix a problem in which trying to set the remote attributes fails. This can occur during a prefetch on an already fetched immutable/appendOnly file.
- This update addresses the following APARs: IV89733 IV89807 IV89899 IV89900.