New in v4

New in v4

New features in v4 described here.

Aggregator can automatically adjust to collection rate changes

Aggregator can automatically adjust to dynamic collection rate changes through update hint in the metadata.

Per-Set Permissions

More control over set access through munge-based per set permissions.

LDMS Metric Set meta data includes metric set update hint and munge-based permissions.

Metric sets seen based on permissions via munge: root (top) sees more than user (bottom).

More control over set Collection and Aggregation

More control over set Collection and Aggregation through:

• Vector of sets - A sampler can collect multiple sets (e.g., vector of sets). This includes storing the same set for multiple timestamps (e.g., for high frequency collection).
• Metric Set Groups - Single RDMA push/pull of multiple disparate sets (e.g., could be meminfo and vmstat). This also supports pulling a vector of sets, so that sets can be aggregated at lower frequency than collection, without losing any data.

A particular benefit is increased reliability of data gathering, even when network events might impede the ability to gather data. Multiple historic metric sets can be retained on-node through such events and then subsequently pulled.

Automatic Failover (HA)

Greater reliability through automatic pairwise failover of aggregators. This replaces the manual triggering of failover that was in v3.

Easier Queries

ldms_ls supports regular expression matching

New Metric Sets

• Cray dvs
• Slurm job id - single job per node (currently)
• Performance counters (perf and papi)

Dynamic Event Support

String and json formatted push-based event handling in the transport via the LDMS_streams interface. This is in contrast to the canonical pull-based mode for metric sets.