Improve leaderless tablet triage and unsafe config change tooling

[pgyogesh]

Summary

• Bump tablet-report-parser.py to v0.60 and make the script executable.
• Replace legacy delete_leaderless_be_careful and UNSAFE_Leader_create views with two structured views for leaderless tablet triage:
  • unsafe_config_change_commands: One row per leaderless tablet with table FQN, replica counts, max-CTID (OpID) peer selection, all peer UUIDs, and a cmd_to_run using /home/yugabyte/tserver/bin/yb-ts-cli.
  • leaderless_peer_ctid_rank: One row per leaderless tablet with per-peer CTID ranking, sync indicators (ctid_in_sync, spreads), lease status, and an aggregated all_peer_ctids column for quick review.
• Add generate-unsafe-config-change.py: Interactive helper that reads unsafe_config_change_commands from a TabletReport SQLite DB and writes a reviewable bash script (unsafe_config_change-run.sh) with readiness prompts, optional TLS certs, and sleep between commands.

Test plan

• Run python3 tablet-report-parser/tablet-report-parser.py <support-bundle/YBA/TabletReport> against a known support bundle.
• Open the generated .sqlite DB and query unsafe_config_change_commands and leaderless_peer_ctid_rank for leaderless tablets.
• Confirm cmd_to_run includes all RUNNING peer UUIDs (not just one peer) and uses the max-CTID tserver.
• Confirm leaderless_peer_ctid_rank ranks peers by (cterm, cidx) and flags out-of-sync CTIDs.
• Run python3 tablet-report-parser/generate-unsafe-config-change.py TabletReport.sqlite, review the generated script, and verify commands match the DB view.

[eugeneckim]

I did a test of this and did "." as my answer and it errored out saying it's a directory, not a file. As a user i saw "script path" and thought it meant the "path." could we make this more clear? perhaps full path and script name?

[pgyogesh]

@eugeneckim At which step did you get this error?

[eugeneckim]

At this precise prompt. i just input the path "." basically the thing expects a file when it says "path" i assumed it was a directory location.

[eugeneckim]

A couple things:

• Curious: Did we go with a separate script for generate-unsafe-config-change.py because it's risky operation and don't want to have it baked into the report parser?
• Could we provide documentation in the README, or perhaps separately on an end-to-end run of this? Similar to the test plan you had outlined.
• Possibly consider providing example files for someone if they want to do a quick test. However, this might need to be internal (e.g. shastra) because it may have internal data.

[pgyogesh]

A couple things:

• Curious: Did we go with a separate script for generate-unsafe-config-change.py because it's risky operation and don't want to have it baked into the report parser?

generate-unsafe-config-change.py is a separate, interactive tool because every deployment differs:

• yb-ts-cli path — install dir varies (/home/yugabyte/..., custom images, non-standard layouts).
• TLS — optional -certs_dir_name and cert directory location per customer.
• Timing — sleep between commands is configurable at generation time.
• Safety — readiness prompts before writing a script; you review the file before bash on a live cluster.

• Could we provide documentation in the README, or perhaps separately on an end-to-end run of this? Similar to the test plan you had outlined.

Updated README.

• Possibly consider providing example files for someone if they want to do a quick test. However, this might need to be internal (e.g. shastra) because it may have internal data.

Will do it in shastra.

[kapil-yb]

@pgyogesh

Are multiple peer UUIDs being passed now? The old view passed only one node_uuid to unsafe_config_change. If so, why is passing all RUNNING peers safer, and when would we actually need this?