Skip to content

Latest commit

 

History

History
39 lines (28 loc) · 1.8 KB

TESTING.md

File metadata and controls

39 lines (28 loc) · 1.8 KB
Raw

Manual integration testing

A basic test setup for slog-agent is like:

  1. Start upstream: run fluentlibtool server provided by fluentlib as a fake Fluentd
  2. Start agent: run BUILD/slog-agent run --config=testdata/config_sample.yml
  3. Feed input: cat testdata/development/*.log > /dev/tcp/127.0.0.1/5140

Current coverage

  • Buffer (HybridBuffer): recovery mechanism
  • Input (Syslog): parsing, error handling
  • Orchestrate (ByKeySetOrchestrator): distribution by key-set (= label sets)
  • Rewrite and Transform: fully covered by unit tests
  • Run: config initialization and reloading

Areas lacking tests

Output: error recovery

The normal code-path for output is already covered, but not any of error related paths that are critical to production environments. Those code are concentrated in:

  • output/baseoutput/clientsession.go
  • output/baseoutput/clientworker.go

Some of the expected errors and special code-paths are:

  • Fluentd keeps receiving chunks but not responding
  • Log chunks sent unsuccessfully
  • Log chunks sent but not acknowledged
  • Interruption in recovery stage before all leftover chunks from the previous session can be sent
  • Sessions reach their max duration and abort for reconnection
  • The need to abort connection when a shutdown signal is received, in order to interrupt any ongoing read/write

Except the session max-duration which can be tested by using very short upstream/maxDuration in config, the rest may be helped by emulation of random network or Fluentd errors in fluentlib and --test_mode of the agent, but there are no automated tests as it'd require not only to verify the final output but also certain code paths being touched and the contents of certain variables at certain stages.