- File uploader
- File manager
- graph-app-kit: Tigergraph support
- pygraphistry: fixes and explicit Python 3.9 support
- File API
- Dask-cuda-worker health check
- Caddy 2
- Mamba: Instead of conda
- Investigator tracking
- Dask-cuda health checks and autohealing
- AWS + Azure now CUDA 11.0
- A100 support and CUDA 11.x variant
- Caddy 2
- File API
- Nutanix vGPU
- Fixes & Tweaks!
- Expose prune isolated
- Clarify launch text
- Work around BlazingSQL concurrency bugs
- RHEL offline installer
|Python PyGraphistry client||0.16.2|
|Caddy||1.0.3 -> 2.3.0|
|Elasticsearch node driver||14.2.2|
|NodeJS||14.15.3 -> 14.15.4|
|Python||3.7.9 -> 3.7.7|
|Neo4j node driver||4.1.1|
|Notebook||6.1.1 -> 6.2.0|
|Splunk node SDK||1.9.0|
|Tornado||5.1.1 -> 6.1|
- Tigergraph examples
- File Uploader:
- Upload arrow, csv, json, orc, parquet, tsv files
- Supports .zip and .tar.gz, and automatically compress before upload if needed
- File Manager: Explore, rename, and delete uploaded files
- File API: Create and manage dedicated files (XLS, Parquet, ...). Create new visualizations from these files without having to reupload data.
- Dask-cuda-worker healthcheck: Internal route /forge-etl-python:8080/workerhealth (public route to be exposed with RAPIDS 0.18/0.19)
- Docker: More configuration hints
- PyGraphistry: Tutorials for using the complex encodings - colors, sizes, icons, and badges
- Nutanix vGPU: Added Nutanix-specific instructions and updated general vGPU instructions
- Tutorial for cURL REST API (incl. File API): Go all the way from JWT auth to uploading data to making a visualization.
Fixes & Tweaks
- Memoization optimization no longer fails with rich values: Will signal a warning and proceed unmemoized, pending upstream Pandas fix / downstream workaround
- Static code analysis is now enforced and many minor style fixes around that
- Python 3.9 added to test matrix
- Expose prune isolated: Expose URL parameter `&pruneOrphans=true`
- Clarify launch text: "Launch visualization" -> "Click to start visualization session"
- Work around BlazingSQL concurrency bugs: Manual concurrency control around BlazingSQL usage. Adds stability both for single-GPU when workers > 1, and for multi-GPU.
- RHEL offline installer: Non-destructive install of docker-compose
- Mamba: To significantly speed up Python library installation, conda has been deprecated in favor of mamba. Notebook users can keep using `pip install --user <package>` as before.
- Python: Version tweaks, unifying around Python 3.7.7 / RAPIDS 0.17 /Notebook 6.2 / Tornado 6.1
- Caddy2: Upgraded Caddy proxy from Caddy 1 to Caddy 2 (2.3.0). This provides performance, feature, and security benefits, including for advanced enterprise environments.
- Investigator tracking: Finer-grainer user authentication has been added to the investigator app (URL /pivot). There should be no visible differences in the current release; this is strictly in preparation for upcoming sharing and cloud features.
- PyGraphistry: The CI/CD/packaging automation layers have been significantly updated; these changes should be transparent
- Dask-cuda autohealing: Service forge-etl-python exposes health checks on dask-cuda-worker (to change in RAPIDS 0.18/0.19), with monitoring by docker-compose.yml
- Azure + AWS CUDA 11.0: The Graphistry containers for Marketplace-launched instances have been upgraded from CUDA 10.2 -> 11.0. This enables all new A100 GPU instance types and remains compatible with all other RAPIDS-capable GPU instance types. The underlying VM Nvidia drivers were already 11.x.
- Graphistry download matrix by CUDA version , incl. A100 GPU support: The downloads page will start releasing two container variants:
- CUDA 10.x - minimal/universal (Typical): Currently 10.2. This works with almost all RAPIDS-capable Nvidia GPUs and is the same container that runs on Graphistry cloud versions.
- CUDA 11.x (mandatory for A100): Required for A100 GPUs to work around an A100-specific issue where Nvidia driver/containers break backwards incompatibility with 10.2. Recommended if specific 11.x optimizations are required, and will switch to recommended default for on-prem users in future versions.
- REMINDER: Host OS CUDA version must be the same or higher than the container version. Ex: 11.x hosts can run 10.x containers, but 10.x hosts cannot run 11.x containers.
- Caddy 2: Custom data/config/Caddyfile configurations must be ported to Caddy 2 format, see Caddyfile.example.
- If you use a custom Caddyfile, port it from Caddy 1 format to Caddy 2 format. See examples in `data/config/` .
- Alternatively, manually defer upgrading Caddy by copying your old `docker-compose.yml` caddy service (section `caddy: ...`) into the corresponding section for the new `docker-compose.yml`, and ensure you have the corresponding caddy image. You will need to repeat the workaround on every upgrade.
- If you perform custom docker automation that hard codes Docker image names, instead of going generically via the docker-compose.yml service names (recommended), image version tags now include CUDA suffix `-10.2`, `-11.0`, and `-universal`. Ex: `graphistry/caddy:v2.35.4-universal`.