tns-mirror-client

Typed, read-only Python access to a tns-mirror database — a local mirror of the IAU Transient Name Server public objects catalogue.

Add the library, get results. No hand-written SQL, and no rate-limited API on your hot path.

$ pip install 'tns-mirror-client>=1,<2'

import os
from tns_mirror_client import TnsMirror

with TnsMirror(dsn=os.environ["TNS_RO_DSN"]) as tns:
    hit = tns.nearest(ra=203.1, dec=10.2, radius_arcsec=3.0)
    if hit:
        print(hit.name, hit.type, hit.redshift, f'{hit.separation_arcsec:.2f}"')

    obj = tns.by_name("AT2026abc")
    print(tns.count(), "objects mirrored")

You need a read-only DSN from whoever operates the mirror. This library never writes, and it carries no TNS credentials — those belong to the server.

Why not just write the SQL?

Because a cone search on a sphere has three edges that are easy to get wrong and fail silently:

• The cosine must be clamped before acos, or an exact positional match — the most common case in a cross-match — raises a domain error.
• Near a pole the RA bound stops meaning anything: two objects 177° apart in right ascension can be 0.2° apart on the sky.
• At the 0/360 seam, ra BETWEEN 359.5 AND 0.5 matches nothing at all.

This library handles all three, and keeps the indexed bounding-box prefilter that makes the query fast. The geometry is tested by sampling the rim of the cone at every bearing and asserting the prefilter never excludes a point that is genuinely inside the radius.

API

TnsObject mirrors the schema columns exactly: objid, name, ra, dec, type, redshift, discoverydate, discoverymag, internal_names, reporting_group, source_snapshot, refreshed_at — plus separation_arcsec, populated only by cone queries.

Everything but objid, name, ra, dec, source_snapshot and refreshed_at is nullable, and a minimal TNS row is common. Do not assume type or redshift is populated; hit.is_classified says whether TNS has published a classification.

Check freshness before you trust it

A row count looks identical whether the mirror synced an hour ago or died three weeks ago:

meta = tns.meta()
if not meta.is_fresh(max_age_hours=26):
    raise RuntimeError(f"TNS mirror is stale: last sync {meta.age} ago")

Versioning

Client 1.x speaks schema v1. The major version tracks the contract, not the project milestone, so pin it:

tns-mirror-client>=1,<2

You are then insulated from additive schema changes — a new nullable column or a new index bumps the minor and never breaks a reader. A breaking change bumps both majors together and is announced in the changelog.

The client reads the mirror's published schema_version on connect and raises SchemaVersionError on a mismatch rather than returning quietly wrong answers. That check needs SELECT on tns_mirror_meta, which the mirror's print-grants output includes; pass check_schema_version=False to skip it.

Connections

TnsMirror(dsn=...) opens lazily and closes with the context manager or .close(). To reuse a connection you already manage — from a pool, say — pass it instead, and the client will not close what it did not open:

tns = TnsMirror(connection=my_conn)

If the mirror was deployed with a non-default TNS_SCHEMA/TNS_TABLE, pass schema= and table=.

The client sets its session read-only as defence in depth, so even a bug in this library cannot write through a privileged role.

Dependencies

One: psycopg. This gets embedded in other people's applications, so results are plain dataclasses rather than a validation framework.

Requires Python 3.11+.

Licence

MIT. The mirrored catalogue itself is not the mirror's to license — follow the TNS data-use terms and cite TNS as they require.