Skip to main content
All tools

What is inside this Iceberg table?

Open a table's metadata.json, or its whole metadata folder. Read the schema history, the snapshots, the manifests and the files they track, with health checks that state their thresholds.

Drop a metadata.json or the metadata folder here.

The files are read in this tab. Nothing is uploaded.

The metadata.json alone
Format version, schema history, partition specs, sort orders, snapshots, branches and tags.
With the .avro files next to it
Manifests of the current snapshot, data and delete files, file sizes, and the health checks that need them.

A table is a tree of small files.

An Iceberg table keeps its own catalogue next to the data. Every commit writes a new metadata.json that holds the schemas, the partition specs and the list of snapshots. Each snapshot points at one manifest list, the manifest list names the manifests, and each manifest lists data files or delete files with their partition values, row counts and sizes.

That is why a table's condition can be read without opening a single data file. Whether snapshots are being expired, how many small files a streaming job has left behind, how much of the table is covered by delete files and whether the manifests need a rewrite are all written in the metadata folder.

The same folder also records how the table got here: every schema change by column id, every change to the partitioning, and the operation behind each snapshot. This tool reads that history and puts it on one page.

Iceberg Inspector FAQ.

What does the Iceberg Inspector read?

The table's metadata.json and, when you add them, the Avro manifest lists and manifests from the same metadata folder. From the metadata file alone it shows the format version, the schema history, partition specs, sort orders, snapshots, branches and tags. With the Avro files it adds the manifests of the current snapshot, the data and delete file counts, file sizes and the health checks that need them. It never opens a data file.

Are my Iceberg metadata files uploaded?

No. The files are read by JavaScript in your browser tab and are not sent to any server. The only analytics the page sends go to analyticslab.ro: the page path, the table's format version, whether manifests were included, whether the files were dropped, chosen or the built-in sample, and whether the contact link was clicked. As with any web request, our server also sees your IP address, your browser's user agent and the site you came from, and stores them in the analytics system we operate. None of it carries a file name, a path, a schema or a table property.

Which health checks does it run, and where do the thresholds come from?

Five. Snapshots older than history.expire.max-snapshot-age-ms, 5 days by default. Data files below 75 percent of write.target-file-size-bytes, 512 MB by default, which is the cut-off rewrite_data_files uses. Deleted rows at 30 percent or more of the rows in data files, the default delete-ratio-threshold of rewrite_data_files. A current snapshot with at least 100 manifests, the default of commit.manifest.min-count-to-merge, that are far below the 8 MB manifest target. A metadata log that has reached write.metadata.previous-versions-max, 100 by default, on a table that does not delete old metadata files after a commit. Each result states the number it used. A table property overrides the default, and where the trigger is our own choice the result says so. Ages are measured from the table's newest snapshot and not from today's date, so the same files always give the same answer.

Which Iceberg format versions are supported?

Format versions 1 and 2 are read in full. Version 3 tables load and their deletion vectors are counted as delete files, but the tool does not decode the contents of a deletion vector, row lineage or the types added in version 3. It lists what it skipped.

Why does it say a manifest could not be read?

Iceberg writes manifests as Avro files, compressed with gzip unless the table sets write.avro.compression-codec. The tool reads uncompressed and gzip-compressed manifests. Manifests written with zstd or snappy are reported as not read, and the checks that need them are marked as not checked.

Which Iceberg version was this checked against?

The results are tested against tables written by Apache Iceberg 1.12.0 on Spark 4.1.4, on 16 November 2026. For five synthetic tables covering format versions 1, 2 and 3, the tool must report the same snapshots, manifests and file counts as Iceberg's own metadata tables.

Built by AnalyticsLab.

We build and repair data platforms: lakehouses on Iceberg, dbt transformation layers and Airflow orchestration. If the health checks found more than you expected, tell us about the table.