Release notes for v2.33.31

Highlights

• Infrastructure:
  • Hypervisor support via RMM GPU memory settings

Graphistry Versions

Server	2.33.NEXT
JS React+Vanilla	3.7.6
Python PyGraphistry client	0.14.0

Third-Party Versions

BlazingSQL	0.16
Caddy	1.0.3
CUDA (In-Docker)	10.1.2
Django	3.0
Docker (CE)	19.03.2
Docker Compose	1.24.1
Elasticsearch node driver	14.2.2
Gremlinpython	3.4.8
Jupyter	1.0.0
Neo4j-python-driver	4.1.0
NodeJS	14.14
Pandas	1.1.2
Postgres	12.3
Python	3.7.7
Neo4j node driver	4.1.1
Notebook	6.1.1
RAPIDS	0.16
Arrow	0.17.1
Splunk node SDK	1.9.0
Tornado	5.1.1

Summary

This upgrade is primarily to add functionality for on-prem enterprise hypervisor users. There is no corresponding cloud marketplace release.

New Features    

• Infrastructure: Hypervisor support via RMM settings:
  • Set: In `data/config/custom.env`,  you  can now  set  RMM GPU memory configurations that get  reused  by  service  forge-etl-python  for  cudf,  cupy,  and  BlazingSQL:
    • RMM_ALLOCATOR: default or managed (default)
    • RMM_POOL: TRUE (default) or FALSE
    • RMM_INITIAL_POOL_SIZE: None or # bytes (default: 33554432 for 32MB)
    • RMM_ENABLE_LOGGING: TRUE or FALSE (default)
  • Inspect: Set `LOG_LEVEL=INFO` and then watch `docker-compose logs -f -t --tail=100 etl-server-python` for the setting used for initializing cudf and other RAPIDS packages
  • Tip: Hypervisors like VMWare vSphere and Nutanix AHV will likely use Nvidia vGPU software, which in turn does not currently support CUDA Unified Memory, which the Graphistry default RAPIDS RMM_ALLOCATOR policy choice of `managed` does not support. The new RMM settings enable hypervisor users to switch to policy `default`.

Migration

No breaking changes; upgrade as usual

• We strongly recommend this upgrade for enterprise users needing hypervisor support, and for on-prem users who are not yet on the previous v2.33.17 release