Release notes for v2.34.14 – Graphistry

Highlights

Core:
- graph-app-kit: quick launch + AWS Neptune
API:
- File Uploader: REST API + PyGraphistry support - Up to 9X faster viz uploads
Infrastructure:
- RAPIDS 0.17
- Dask support: multicore + multi-GPU (early)
- Postgres upgrade
- More autorestarts
Administration
- RHEL 8.x: Environment bootstrap, including air-gapped
- nvidia/cuda base container: For easier testing
- Initial user flow: Service account creation script + new AWS flow
Docs:
- PyGraphistry: Complex encodings tutorials
Fixes & Tweaks!
- Filters: Dates and lists; Nulls in like/ilike expressions
- Neo4j 3.x connector: Text indexes
- Dask: Initialization

Graphistry Versions

Third-Party Versions

Summary

New Features

graph-app-kit:
- one-click launchers for AWS: minimal + full
- Improved AWS Neptune support: More demos, documentation, and quick launchers
dask & dask_cudf notebook support

API

File API
- REST API: Contact for early documentation
- PyGraphistry: See https://github.com/graphistry/pygraphistry/pull/195
  - ~9X faster plot calls for the upload step when node/edge data has been used before in the same session
  - Start using via `.plot(as_files=True)` under `graphistry.register(api=3)` mode

Docs

PyGraphistry: Tutorials for using the complex encodings - colors, sizes, icons, and badges

Fixes & Tweaks

Faster pageload
'75%' page load bug fixes: Fewer incidents and better auto-recovery
Faster 'move node' action
Filters:
- Speedups
- When manually inputing filters via the UI, corrected value types inference
- When manually inputing filters via the UI, corrected "IN ( ... )" 1-item list parsing
- LIKE/ILIKE expressions over columns with mixed string+null values now works
Neo4j 3.x connector: Text indexes now analyzed (regression fix)

Infrastructure

GPU Dask notebooks: Graphistry now supports dask / dask_cudf, including from within notebook:
```
import cudf, dask_cudf
from dask.distributed import Client
with Client('dask-scheduler:8786'):
  gdf = cudf.DataFrame({'x': [1,2,3]})
  dgdf = dask_cudf.from_cudf(gdf, npartitions=2)
  dgdf['x'].sum().compute()
```
Note the use of `with` statement to ensure the client and its resources get deallocated after use, otherwise you risk leaking GPU memory
GPU Dask + BSQL: Multi-GPU & bigger-than-memory
- Graphistry internal use is Initially aimed towards single node for larger-than-memory file uploads and fewer blocking server calls
- The initial rollout uses the following services for single-node multiple-gpu:
  - docker service dask-scheduler
  - docker service dask-cuda worker, which manages access to all locally available GPUs
  - docker service forge-etl-python, which is multiple python processes
    - transitioning from in-process single-gpu to dask-based multi-gpu
  - existing service streamgl-gpu, which remains single-gpu
- See configuration options in data/config/custom.env.template
Dependency upgrades: Minor version patches, including Postgres, Python stack, Node, and web servers
More autorestarts: The streamgl-viz and streamgl-gpu services will more aggressively restart after visualization initialization failures

Administration

RHEL 8.x: Reference bootstrap scripts and instructions in etc/scripts/bootstrap/rhel8
- Installation tested with RHEL 8.1-8.3, docker 19.03, Nvidia driver 455, CUDA 10.2/11.0
- RedHat does not officially support Nvidia (due to epel) nor Docker (due to its competing Podman); script instead uses community-standard non-proprietary centos 7/8 procedures
- Included is an air-gapped guide that splits into offline-download.sh (run from an internet-connected environment clone) vs. offline-install.sh (scan & import download data, then install without needing internet access) and initial hypervisor setup guidance
- Azure RHEL requires fixing Microsoft's dnf repository misconfigurations; see readme for lines to uncomment
nvidia/cuda base container: Container `nvidia/cuda` is now explicitly loaded and exposed as part of distributions, e.g., during `docker load -i container.tar` manual installs. This should simplify environment testing of Nvidia/Docker dependency setup.
Initial user flow
- Optional script etc/scripts/graphistry-service-account.sh for scripting automatic creation of initial admin user
- AWS: Comply with Marketplace secure boot requirements via initial user with credentials admin / your-instance-id

Migration

No breaking changes; upgrade as usual

The major version increase is due to the database version change and introduction of dask_cudf

If an AWS user, an initial default user now gets autopopulated: admin / the-instance-id