cs-codex-dist-tests/Tests/CodexContinuousTests/reports/CodexTestNetReport-July2023.md

3.8 KiB

Codex Continuous Test-net Report

Date: 02-08-2023

Report for: 07-2023

Test-net Status

  • Start of month: Offline - faulted
  • End of month: Offline - faulted

(Faulted: Tests fail with such frequency that the information gathered does not justify the cost of leaving the test-net running.)

Deployment Configuration

Continuous Test-net is deployed to the kubernetes cluster with the following configuration:

5x Codex Nodes:

  • Log-level: Trace
  • Storage quota: 2048 MB
  • Storage sell: 1024 MB
  • Min price: 1024
  • Max collateral: 1024
  • Max duration: 3600000 seconds
  • Block-TTL: 120 seconds

3 of these 5 nodes have:

  • Validator: true

Kubernetes namespace: 'codex-continuous-tests'

Test Overview

Changes Test Description Status Results
New in 07-2023 Two-client test Every 30 seconds, two nodes are chosen at random. A 10 MB file is generated and uploaded to one. 10 seconds later, it is downloaded from the other. File contents are asserted to be equal. Faulted Test reliably fails after 30 to 45 minutes. Both upload and download failures occur.
New in 07-2023 Transient-node test Every 1 minute, a new, transient Codex node is started and bootstrapped against a random node of the test net. A 10 MB file is generated and uploaded to the transient node. The file is then downloaded from a second random test net node and file equality is asserted. After that, the transient node is shut down. 30 seconds later, a new transient Codex node is started and bootstrapped against a third (guaranteed different from first and second) node from the test net. The same file is then downloaded from the new transient node. File equality is asserted. Not running Test was not run because the previous more rudamentory test has faulted.

Action Points

  • Codex logs (from these long-running containers) are often incomplete when downloaded after a test failure. A reliable way of maintaining these logs is needed. The logs can quickly explode in size, even for test-nets that run only for a few hours.
  • The distributed testing setup can and has been used to reproduce the failure of the Two-client test in a local environment. Investigation is on-going.