solana/net
Michael Vines 017eb10e76 Add file header doc 2018-09-05 09:02:02 -07:00
..
remote Add file header doc 2018-09-05 09:02:02 -07:00
.gitignore Morph gce_multinode-based scripts into net/ 2018-09-05 09:02:02 -07:00
README.md Add Tips section 2018-09-05 09:02:02 -07:00
common.sh Private IP networks now work, and are the default 2018-09-05 09:02:02 -07:00
gce.sh Work around concurrent |gcloud compute ssh| terminal issue 2018-09-05 09:02:02 -07:00
init-metrics.sh net/ can now deploy Snaps 2018-09-05 09:02:02 -07:00
net.sh Enable cargo features to be specified 2018-09-05 09:02:02 -07:00
ssh.sh Morph gce_multinode-based scripts into net/ 2018-09-05 09:02:02 -07:00

README.md

Network Management

This directory contains scripts useful for working with a test network. It's intended to be both dev and CD friendly.

User Account Prerequisites

Log in to GCP with:

$ gcloud auth login

Also ensure that $(whoami) is the name of an InfluxDB user account with enough access to create a new database.

Quick Start

$ cd net/
$ ./gce.sh create -n 5 -c 1  #<-- Create a GCE testnet with 5 validators, 1 client (billing starts here)
$ ./init-metrics $(whoami)   #<-- Configure a metrics database for the testnet
$ ./net.sh start             #<-- Deploy the network from the local workspace
$ ./ssh.sh                   #<-- Details on how to ssh into any testnet node
$ ./gce.sh delete            #<-- Dispose of the network (billing stops here)

Tips

Running the network over public IP addresses

By default private IP addresses are used with all instances in the same availability zone to avoid GCE network engress charges. However to run the network over public IP addresses:

$ ./gce create -P ...

Deploying a Snap-based network

To deploy the latest pre-build edge channel Snap (ie, latest from the master branch), once the testnet has been created run:

$ ./net start -s edge

Enabling CUDA

First ensure the network instances are created with GPU enabled:

$ ./gce.sh create -g ...

If deploying a Snap-based network nothing further is required, as GPU presence is detected at runtime and the CUDA build is auto selected.

If deploying a locally-built network, first run ./fetch-perf-libs.sh then ensure the cuda feature is specified at network start:

$ ./net.sh start -f "cuda,erasure"