Versions Compared

Version Old Version 8 New Version 9
Changes made by

Anjali Tyagi

Bala Harish A(AC)

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

  • OSS Versions — See [DRAFT] Third-Party Components for Storage 15.3 for the complete listing of packages and versions for this release.

  • Fixed in 15.3

    • Improved stability for dense clusters — Improved CPU utilization in dense clusters which reduces spontaneous process restarts and other instabilities. (CUP-630)

    • Improved handling of S3 multi-deletes — Addressed a multi-delete ES ghost entry bug impacting various Veeam-using customers. (SWAR-9703) 

    • EC multipart & versioned segments — Fixed the source of data loss involving missing segments from obsolete EC object versions. (SWAR-9770)

    • Finalizing state on restart/shutdown — Fixed an issue where a node does not finalize the state during restart or shutdown. Instead, the node requires manual power for cycling. (SWAR-9767)

    • Network disruption — Fixed the source of data loss when a cluster had significant network disruption for multiple HP cycles. (SWAR-9753)

    • EC conversion — Fixed a rare cause of data loss with EC conversions interleaved with volume recoveries. (SWAR-9751)

    • Elasticsearch upgrade — Swarm Storage 15.3 requires upgrading Elasticsearch to 7.17.9. The rolling upgrade of the Elasticsearch cluster can be done either before or after upgrading to storage 15.3. (SWAR-9731)

    • cip.processes — The default cip.processes value is changed to 2. (SWAR-9841)

Watch Items and Known Issues

...

  • Several settings that are persisted in the stream of cluster’s persistent settings have their defaults changed in this version. It is recommended to review these settings and make appropriate adjustments.

    • health.examDelay now defaults to 0.19. It is recommended Not to use a value lower than the default.

    • health.fvrPushDelay has a new default of 0.7, which is recommended for most clusters.

    • power.savingMode has a new default of False, which is recommended for most clusters.

    • scsp.defaultSynchronousIndexWait has a new default of 60, which is recommended for most clusters. This is a non-persisted setting.

  • Configuring elasticsearch.yml's network.host (https://www.elastic.co/guide/en/elasticsearch/reference/7.16/important-settings.html#network.host) to "__site__" might not choose the right IP to allow master election if the server is multi-homed. Modify the elasticsearch.yml to enter a specific IP for the node and the configuration script will preserve it. (SWAR-9350)
    If you run into this issue, the fix is to:

    • systemctl stop elasticsearch on all ES nodes

    • remove all the contents of the path.data directory

    • change network.host: <IP of ES NIC in the Storage VLAN>

    • systemctl start elasticsearch

  • Elasticsearch can fail to start and return a warning "unable to load JNA native support library", which is due to SELinux setting “noexec” on /tmp.
    For Elasticsearch 7.5.2, edit “/etc/elasticsearch/jvm.options” replacing the line "-Djava.io.tmpdir=${ES_TMPDIR}" with "-Djava.io.tmpdir=/var/log/elasticsearch". With Elasticsearch 7.17, uncomment the "Environment=ES_TMPDIR=/usr/share/elasticsearch/tmp" line in /etc/systemd/system/elasticsearch.service.d/override.conf and create that directory. (SWAR-9347)

  • Swarm versions 10.0 onward are vulnerable to kernel issues manifested on some Intel CPUs. Symptoms include lowered performance, long mount times, and cluster instability. Swarm versions 14.1 and later provide a workaround for this issue, see https://caringo.atlassian.net/wiki/spaces/KB/pages/2973204604. (SWAR-9055)

  • Customers who perform paginated listing queries (using sort and marker) need to choose a unique set of fields to return the complete results. (SWAR-9630)

  • Writes of objects into versioned buckets suffer a small (constant) performance penalty. (SWAR-9794)

These are standing operational limitations:

...