TODO.md

TODO

curses_helper.py

• We need to rewrite the UI to remove reliance on stdscr; stdscr should just be blank canvas. This solves all the rescaling issues, as well as limitations with what characters can go where, etc.
• The timestamp needs to be its own window. The border should be part of the bottom-most canvas and only needs redrawing if we resize the window or toggle borders.
• Scrollbars need their own windows.

All

• listgetters need to support passing both a label selector and a match selector in listgetter_args
• When running either cmt or cmu, check whether .ssh/id_ecdsa.pub is in authorized_keys in .cmt/ansible/inventory.yaml; if not, add it.
• Rewrite command_parser to treat options passed before a command as global, and to allow options interspersed with arguments. Perhaps even add short options.
• Add --dry-run support for more commands.
• Is it possible to rewrite the generator/processor system in a way that processors could be completely eliminated?
• Introduce .kube/current-context and have all clusters in .kube have their own files (named config-clustername) rather than merging the config-files

cmu

• Divide ping (inventory & context views) into 2 different categories; normal and slow; slow to use with hosts that responded with "NO ROUTE TO HOST", "CONNECTION TIMED OUT", or "COULD NOT RESOLVE"; we still continue pinging them every, say, 30 seconds, but there's no point holding up the results from the remaining hosts. Also we should probably figure out massively parallel pinging somehow.
• Bundle all Core APIs into one file and load them all using secure_read_yaml_all(); this should lead to a slight performance improvement; note that the files should be kept separate and only bundle upon "build"
• In Node and Inventory Info view we should provide a way to list all logs related to the host, and a shortcut to jump to that log.
• Try to find other things we can simplify in the views.
• Move field_templates and built-in views to views.
• Modify generator_list; ideally every list element should be typed; this way we'd be able to use, for instance: [(kind, api_family), address, address, ...] and have the (kind, api_family) tuple formatter by that kind of parser, while the addresses are formatter by the address formatter, without having to special case various types of lists. We might even want the list of types be possible to generate on the fly; currently it can only be specified manually. Note: this would need quite a bit of a rewrite, but it would probably be healthy and hopefully cut down on the amount of special cases.
• Audit and make a list of all necessary types
• Make generic_infogetter consistent WRT to paths: ["literal", ["path"], [["alternate1", "alternate2"]]]
• All use of curses should be abstracted away; ideally cmu should be able to use a variety of different toolkits to present its data
• All generators and other functions used to extract data should be removed from cmu; that way we don't have to duplicate functionality between cmu and cmt

cmtinv:

• Optionally limit rebuild-inventory to a subset of clusters.

cmtadm:

• Add command to import kube-config (requires cluster-name--unless unique) and a path.
• prepare_passwordless_ansible won't work on localhost; we're not passing the password, and the password might not be the same on the remote system and the local system anyway.
• Add pre-upgrade-check that checks whether relevant config files (notably containerd) are compatible with new settings. Also check whether the cluster currently uses any APIs that are deprecated, and any deployments that are known to be unsupported on newer versions of Kubernetes. Config-checks needs to check all nodes, not just the control plane.
• troubleshoot:
  • Check that the config-files as installed/created at install time by cmtadm/cmt are in sync with what's on the machines in the cluster at present, and if not, warn about the difference. The differences may be intentional, so we cannot indiscriminately overwrite them, but the user should at least be aware of the differences (we could even show a diff if --verbose is passed) and provide a helpful message about what playbook to run to update the files.
  • Add a security warning about file permissions.
  • Add a security warning ansible_pass being used in cmtconfig.

logparser:

• All handling of message needs to be rewritten; as soon as a message gets formatted as a themearray it can no longer be processed further.
• Rewrite key_value; the parser is overly complex at the moment; it needs to be simplified and have fewer special cases.
• Line too long shouldn't override severity. | Ideally max line length should be a.) configurable, b.) trigger line splitting into remnants
• We cannot replace tabs with spaces in the logparser; we need to do it in the printers instead; this way we know the real line length (due to facility etc. we might not have the same starting point for every line, so expanding tabs into spaces won't work properly). | Currently we strip tabs; if we want to handle them we need to modify cmtlib.py:split_msg()

kubernetes_helper:

• Replace playbooks/drain_node.yaml with cordon_node() + post evictions (or delete if PodDisruptionBudget causes issues) for all non-DaemonSet pods.