Release notes for v2.27.8 – Graphistry

For further information, see the release blogpost.

Graphistry Versions

Server	2.27.8
JS React+Vanilla	3.7.1
Python PyGraphistry client	0.10.4 (from 0.9.69)

Third-Party Versions

BlazingSQL	0.11 (from 0.43)
Caddy	1.0.3 (from 1.0.1)
CUDA (In-Docker)	10.0
Docker (CE)	19.03.2
Docker Compose	1.24.1
Elasticsearch node driver	14.2.2
Pandas	0.24.2
Python	3.7.3
Neo4j node driver	1.7.6
RAPIDS	0.11 (from 0.10)
Arrow	0.15.1 (from 0.14.1)
Splunk node SDK	1.9.0

Summary

The primary updates are around RAPIDS 0.11 support, more settings for low/high capacity, fixes, and improved pageload speed.

New Features

Core
- RAPIDS 0.11 : ML, out-of-memory, perf/fixes
- BlazingSQL 0.11 : More SQL coverage, Hive
- Faster page load on bigger graphs (and more to come!)
- Clustering auto-stops after 1min, decreasing the risk of service degradation for other users
  - Override by setting GRAPH_PLAY_TIMEOUTMS to 0 for no timeout or the preferred number of milliseconds
- Control number of network streaming and light analytics workers with environment variable STREAMGL_CPU_NUM_WORKERS - defaults to `max` (number of CPU hyperthreads)
- Control number of upload node processes with environment variable
  
  PM2_MAX_WORKERS - recommended to be "max" (number of CPU hyperthreads)
Notebooks/PyGraphistry
- nodexl support:
  - graphistry.nodexl("https://file.xls").plot()
  - graphistry.nodexl("/usr/file.xls").plot()
  - options verbose=True, source="twitter"/"mediawiki", and engine=(pandas Excel engine name)
- zip, neo4j driver, neworkx, igraph, and nb_black preinstalled
  - Neo4j official tutorial updated
- Regular testing now on both Python 3.6 + 3.7
- Pandas 1.x support (built-in is still 0.25.x)
Infrastructure
- Cross-server migration: Faster, handles more data (tested with 30GB+), and is resumable/restartable
  - See updated `/etc/scriptsmigrate.sh` header for new options
- On-prem same-server migration/upgrade script:
  - Same-server variant of cross-server migrate.sh
  - See `etc/scripts/local-migrate.sh`
  - FROM_PATH="/var/g_old" ./migrate-local.sh
  - FROM_PATH="/var/g_old" DOCKER_SUDO=sudo ./migrate-local.sh
  - FROM_PATH="/var/g_old" TO_PATH="/var/g_new" ./migrate-local.sh

Docs

Live Python API registration example now included in the user account dashboard pillbox
List of built-in types for when configuring new data sources (also logged upon system start): https://github.com/graphistry/graphistry-cli/blob/master/docs/configure-ontology.md
RAPIDS multi-GPU deployment planning

Fixes & Tweaks

UI & Viz
- Layout defaults: Investigations disable dissuade hubs and prune-isolated by default
- Home screen's Python API key demo: fixed incorrect parameter "api" to "key"
- Histograms and filtering: Handling of small values between 0 and 1
- Data brush: Histogram and inspector highlights work again
- More reliable loading of preset encodings (color, size, icon, ...) during pageload
Infra
- Docker-based logger tags lines with container context
- AMIs: Ubuntu's snapd service disabled by default, fixing sporadic CPU/memory denial of services
- Caddy: Disabled default-on telemetry
Neo4j:
- Neo4j schema inference handles the case of a partially-populated database, rather than failing and preventing connector load. To work around in older versions, set inferSchema to false or make sure your database is populated.
- When Neo4j schemas use Graphistry-reserved values, prevent those values from corrupting the visualization by dropping them

Migration

We recommend most 2.X users to upgrade due to RAPIDS stability improvements and other bugfixes

Optional (not recommended): Set GRAPH_PLAY_TIMEOUTMS=0 in custom.env to preserve old behavior of clustering with no timeout
Warning - igraph dependency not included as official builds are currently broken
Warning - one-way upgrade: The Arrow upgrade means new (0.15.0+) generated Arrow files cannot be read by older versions of Graphistry, meaning it will be difficult to uninstall the new version of Graphistry and go back to 2.26 without manual data conversion of `data/*/*.arrow` files. For similar reasons, we recommend testing the new version on old files before going live.
Warning: PyGraphistry 0.10.0 drops Python 2.7 support due to Python 2.7 being now beyond EOL.