Where it runs is your decision
Both options use the same binary and the same configuration file.- In your network
- In a cloud account you own
A host or a container in the datacentre, beside the data.Database traffic stays on the LAN. You operate one more thing in the
datacentre — patching, logging, monitoring.Choose it when the datacentre is where your operations already are.
What TigerGate learns about a site
Deliberately little, and worth being able to state to an auditor:- That the site exists, its name, and when it last checked in
- How many targets it reports, and of what kind
- The verdicts of the scans it ran
Set one up
Register the site
Register the site through the API, naming it after the NETWORK rather than a
database — findings are attributed to the site, and one connector in
Frankfurt and one in Virginia have to be tellable apart.The response carries
connector_id and a tgconn_ secret. The secret is
returned once — it is hashed on save and cannot be recovered. If you lose
it, register a new site and revoke the old one.Describe the targets
List each database or share in
connector.json: an id, a kind, a host, a
port and the tiers it may be scanned at. A target registered for metadata
only will refuse a content job even if the platform asks for one, so depth
is enforced at the edge you control rather than only in our cloud.This file carries no passwords and is safe to commit.Supply the credentials
One environment variable per target, named after the target id — uppercased,
with non-alphanumerics as underscores:The connector refuses to start when a target has no credential, and names
the target. The single hyphen is how no password on purpose stays
distinguishable from forgot to set it.
Probe before you schedule
Supported targets
| Kind | Notes |
|---|---|
| PostgreSQL, MySQL, MariaDB | TABLESAMPLE on PostgreSQL; MySQL has no equivalent, so it samples with LIMIT. |
| SQL Server, Oracle, SAP HANA | Read-only user with SELECT on the schemas in scope. |
| MongoDB | $sample, secondary-preferred reads, nested documents flattened. |
| Trino, Presto, Hive | Hive is reached through Trino rather than HiveServer2. |
| Apache Kafka | Samples from the end of every partition, so the current schema is what gets seen. |
| SMB / CIFS shares | NTLM authentication. Recycle-bin and snapshot trees are skipped. |
| NFS v3 | Connector only. NFSv3’s AUTH_UNIX means the client asserts its own identity, which is a property of the protocol — exposing an export to our cloud would mean exposing it to the internet. |
NFS v4 with Kerberos is not supported. It is a different security model rather
than a variation on v3, and treating it as one would misrepresent what the
connector authenticates as.