Release notes for v2.41.3

Overview

This major release updates Graphistry core server dependencies CUDA to 11.8 and RAPIDS to 23.10, which is a breaking change. Many changes have been made to the GFQL endpoint to improve its usability and security, as well as add a few new features. Another new feature, the Python endpoint has also been added, allowing arbitrary Python code to be used in dataset processing. On the UI side, a new button has been added to the data table which allows node and edge files for datasets to be downloaded in various formats. A few smaller issues were also addressed.

NOTE: This bundles and extends patches from hotfix release 2.40.75

Graphistry Versions

Server 2.41.3 (was 2.40.74)

JS React+Vanilla 5.1.1

Python PyGraphistry client 0.34.1 (was 0.33.8)

Third-Party Versions

Arrow 10.0.1

Caddy 2.7.6

CUDA (In-Docker) 11.8 (was 11.5)

Dask 2023.1.1

DGL 1.1.1

Django 3.2

Docker (CE) 23.0.1

Docker Compose 2.17.3

fsspec 2023.1.0

Gremlinpython 3.5.0

Jupyter 1.0.0

Neo4j-python-driver 4.4.5

Nginx               1.25.4

NodeJS 14.21.3

Pandas 1.5.3

Postgres 14.11

Python 3.8.16

neo4j node driver 4.0, 4.1, 4.4

networkX 2.6.3

notebook 6.5.2

Nvidia driver (cloud) 550 (was 470)

RAPIDS 23.10 (was 23.02)

Redis 6.2.7

Splunk node SDK 1.9.0

Tornado 6.2

torch 1.12.1

torch-geometric 2.3.1

torch-god 0.3.1

Features

• Add a download button to the Data Table for downloading node and edge datasets in json, parquet, and csv formats.
• Add play functionality as an aid for viewing and interacting with data in the timebar.
• Add an option to receive GFQL result data in parquet or csv formats.
• Add an option to receive only node or edge GFQL results.
• Add an option to receive only a subset of node or edge properties included in GFQL results.
• Add an option to change serialization options for GFQL results.
• Add an endpoint to allow data processing to be done server-side by executing user provided Python code. This endpoint is gated with feature flags within Graphistry and environment variables.
• PyGraphistry:
  • Added ring layouts `ring_categorical_layout()`, `ring_continuous_layout()`, `time_ring_layout()`
  • Plottable interface includes `encode_axis()`, `settings()`
  • New transform on edge weights `modularity_weighted_layout()`
  • CuDF fixes
  • Added `personalized_pagerank` to the list of supported `compute_igraph` algorithms.

Infrastructure

• Expose telemetry services through caddy and simplify telemetry configuration.

Admin

Security

• Fix GFQL endpoint incorrectly using internal caller.
• Allow GFQL endpoint to accept CSRF tokens as a method of request authentication.

Fixes & tweaks

• Fix special characters in JSON properties are not escaped when getting files and datasets.
• Fix timebar histogram binning incorrectly.
• Fix datetime handling due to RAPIDS 23.10 changing the default policy.
• Fix categorical encodings not working when applied to continuous data.
• Fix some datasets not working correctly when calling GFQL or Python endpoint.

Docs

• Various GFQL documentation fixes.
• Added Python endpoint API documentation.

Migration

• Host Nvidia driver version must now be CUDA 11.8+ (vs 11.5 previously) or 12+.
• They are now rootless, but we keep using root user inside the Graphistry containers. We do not expect breaking changes, but note the significant change in case advanced integrations such as around migration automation scripts are impacted.
• Azure Marketplace images disable NVIDIA GPUDirect Storage (GDS) by default, Enable manually by removing `LIBCUDF_CUFILE_POLICY=OFF` from your `$GRAPHISTRY_HOME/.env` and restarting (force-recreate) all GPU containers.
• (pygraphistry) Dropping support for Python 3.7 (EOL).
• See previous notes.