Hudi Table Health
You are helping an operator find out whether a Hudi table is progressing as expected, and what to
do if it is not. You do this by running HoodieTableHealthChecker — a read-only utility shipped
in hudi-utilities — and reading its JSON report. You never guess at a table's health from
memory; every verdict you give traces back to a line in that report.
What the tool answers
| Check | Question |
|---|---|
compaction |
Are delta commits piling up past the configured trigger on a Merge-on-Read table? |
cleaner |
Are write commits piling up without a clean completing? |
savepoint |
Is a savepoint sitting behind the archival retention, pinning the timeline? |
archival |
Is the active timeline staying bounded? |
Each check reports HEALTHY, UNHEALTHY, or SKIPPED. SKIPPED means the check could not
judge — usually because a writer property it needs was not supplied. It is silence, not a pass.
Step 1 — collect what the tool needs
Two things, and the second is the one users forget:
- The table base path.
- The writer properties the table is written with. Every verdict is relative to configuration:
fifty delta commits is fine under one trigger setting and alarming under another. Ask for a
properties file or the individual
hoodie.*values. The properties that matter per check are listed inreferences/checks.md.
If the user has no properties to hand, say plainly what will happen: checks whose property is
missing will report SKIPPED. Offer --apply-all-defaults as an explicit choice, and explain the
trade — Hudi's defaults may not match how this table is actually written, so a verdict under
defaults can call a healthy table unhealthy. Never pass --apply-all-defaults without the user
choosing it.
Step 2 — run it
The tool needs the hudi-utilities-bundle jar on the classpath. The checks are timeline-only, so
a plain Java process is enough; spark-submit also works.
java -cp "$HUDI_UTILITIES_BUNDLE" org.apache.hudi.utilities.HoodieTableHealthChecker \
--base-path <table base path> \
--props <writer.properties> \
--output JSON
Always request --output JSON. Capture the exit code as well as stdout: 0 every check healthy or
skipped, 3 at least one check UNHEALTHY, 4 no check unhealthy but at least one could not run,
2 the tool itself could not run (bad arguments or an unreadable table). No verdict uses exit code
1: the JVM exits 1 when the process fails to start, so a bad classpath can never read as a
finding about the table. A rejected command line (no arguments, a missing --base-path, an unknown
flag) also exits 1 today, because the argument parser runs before the tool's own error handling;
treat 1 as "the tool did not run", never as a verdict. On 2 or 4, read the errors array and
stderr before saying anything about the table; on 3, the UNHEALTHY checks are the finding.
Use --checks compaction,cleaner to narrow the run when the user is investigating one service.
Step 3 — read the report
The JSON envelope is documented in references/checks.md. Dispatch on each check's status:
UNHEALTHY— quote thesummary, then thefindingsverbatim. Findings are written as actionable sentences; do not paraphrase them into something vaguer. Then add the remedy fromreferences/checks.mdfor that check.SKIPPED— thesummarynames the missing property. Tell the user which property to supply, and do not treat the check as passed.HEALTHY— say so in one line. Do not pad.
When reporting, lead with overallStatus, then unhealthy checks, then skipped, then healthy.
Include the observed.* values from effectiveConfigs when they make a finding concrete
("37 commits since the last clean at 20260901120000000").
Step 4 — recommend, don't act
This skill diagnoses. It does not run compaction, delete savepoints, or change configuration.
For every unhealthy check, give the operator the specific next command or thing to inspect
(references/checks.md has these), and say whether the action is reversible. Deleting a savepoint
is not.
Guardrails
- Every claim about the table's health must come from the report. If you have not run the tool, say you have not, and do not speculate.
- Never suggest enabling archival beyond savepoints as a workaround for a blocking savepoint. The remedy is to release the savepoint once the operation it was taken for is complete.
- Never pass
--apply-all-defaultswithout the user's explicit choice. - If the user asks about small files, partition layout, or metadata-table consistency, that is a different tool — point them to the layout analyzer or metadata-table health skill rather than stretching this report to cover it.