Warden

Warden
/ˈwɔːd(ə)n/
noun

a person responsible for the supervision of a particular place or activity
or for enforcing the regulations associated with it.

Warden is the gatekeeper for data entering our system.

Metadata

The result of warden runs will be stored as metadata in files shadowing the actual data, or in the metadata store. From this metadata the following state will be clear:

• The state of the dataset:

  • passed: all checks complete and ok
  • failed: some checks failed

• Statistics about this feed. These statistics will be enough to do over-time-comparisons etc...

• List of checks, with results, run against this dataset.

• Enough to do roll-ups to determine the "health" of a feed overall.

• These should be something we can eventually push to a centralized service (brandix?).

Running

Warden is a command line tool, and is run directly against a local copy of a view.

Where a view is the traditional hive-style, year=..../month=../day=../dataset partitioning.

Warden runs on both the view-level and the file-level. View checks are run on a complete view, or any nonempty subset of a view; these consist of row counts, numeric summaries, et cetera and can be schema-aware. File checks perform very basic sanity checks at the file level (e.g., does the file have a nonzero size).

Metadata from view checks is to be stored in the metadata store, e.g., s3://ambiata-prod-live-state/$customer/warden/$view. Metadata from file checks is to be stored back into the view, with the naming schema _${view_filename}.warden. However, warden does not write to S3 itself - it writes metadata to the local filesystem, which can then be stored in S3 using other tools.

This means you should be always able to run the following on a configured view:

# basic file checks
warden sanity path/to/view
# full view checks
warden check path/to/view

# infer schema from check metadata
warden infer _warden/$view/$data_dates/$check_date/*.json

# validate the schema which was generated
warden schema validate schema.json

# locate failed check markers
find _warden -name \*.warden -print0 | xargs -0 warden marker failed

Extracting numeric data

Various numeric data can be extracted as CSV files from view markers for easier interactive analysis.

Numeric samples

warden-sample extract \
    -o samples.csv \
    _warden/$view/$data_dates/*/*.json

Summary statistics

warden-sample summarise \
    -o /tmp/stats/$view \
    _warden/$view/$data_dates/*/*.json

Future Work

• Tighter integration with monitoring tools (sending alerts, feeding into smarter online checking etc...).
• warden integration for downstream tools so they only accept warden approved feeds.

Further reading

• Documentation (or find a built version on S3 under s3://ambiata-dispensary-v2/doc/master/warden/).
• Talk slides.