rapid7 / cassandra-utils Goto Github PK

In v0.3.2, the token_cache method creates a new File in create, read, write, truncate mode. This means the file is always "empty" when cached_tokens is called. Instead, we should be creating the file with a custom mode, and caching the File object.

def token_cache
  mode = File::CREAT | File::RDWR | File::SYNC
  @token_cache ||= File.new(token_cache_path, mode)
end

That will create the file if it doesn't already exist, open the file for reading and writing, and allow us to eliminate the flush call when writing tokens.

We also need to call File.seek before reading tokens so that we read them from the beginning of the file.

token_cache.seek 0
token_cache.read

And we need to call File.seek and File.truncate before writing tokens so we overwrite any existing data in the file.

token_cache.seek 0
token_cache.truncate 0
token_cache.write data.to_json

cassandra-utils runs both stats and autoclean (automatic cleanup scheduling). It'd be nice to allow users to run stats and autoclean independently if desired. This could either be a series of command line switches e.g. --no-autoclean and --no-stats or a config file with a list of tasks to run.

Currently the autoclean code uses nodetool ring to get the list of tokens the node owns. This returns the token list for all nodes in the ring, which is kind of an expensive cal. We should use nodetool info --tokens instead.

There's a bug in v0.3.1-beta.1 where we'll start nodetool cleanup even if we've already found an existing cleanup process running. We should add a regression test to capture this bug when we fix it.

We're caching the Mixlib::ShellOut instance when running commands. This has bitten us before (see PR #33 in daemon_runner). We should either skip the caching or just change this to use daemon_runner.

In version 0.3.1-beta-2 the compaction and cleanup stats are no longer being reported.

This was broken in PR #18 when Mixlib::ShellOut was changed to DaemonRunner::ShellOut. The Cassandra::Utils::CLI::Base#run! method wasn't changed, so it's still trying to call Mixlib::ShellOut#run_command and Mixlib::ShellOut#error! instead of DaemonRunner::ShellOut#run!.

Interestingly, there don't appear to be any errors or warnings in the logs about missing method calls. We should look into what's eating those exceptions.

When the autoclean task saves a list of tokens, it will always have version 0.0.1. The code that returns the version is looking for a VERSION file. We don't have a VERSION file, since we're using thor-scmversion and using Github tags.

We should update our CI process to create a VERSION file that gets packaged into the gem when it's built.