aidp-exacs

name	aidp-exacs
description	Read or write Oracle Exadata Cloud Service (ExaCS) from an AIDP notebook via the AIDP `aidataplatform` Spark format handler. Use when the user mentions ExaCS, Exadata, Exadata Cloud, RAC SCAN listener, or has a private-subnet Oracle DB. Prefer the official `ORACLE_EXADATA` connector sample with catalog.id, pushdown.sql, and connector write modes.
allowed-tools	Read, Write, Edit, Bash

aidp-exacs — Oracle Exadata Cloud Service via AIDP aidataplatform

When to use

• User wants to read or write an ExaCS PDB from an AIDP notebook.
• User mentions: "ExaCS", "Exadata Cloud", "RAC SCAN listener", "private-subnet Oracle DB".

When NOT to use

• For Oracle AI Lakehouse / ADW / ATP (Autonomous DB family) → use aidp-alh.

Critical infrastructure prerequisites

ExaCS sits in a customer private subnet; AIDP runs in its own VCN. Two pieces must be in place before any JDBC will work — neither is configurable from a notebook:

1. VCN routing AIDP→ExaCS (PE/RCE through PAGW). Workspaces created without private connectivity can't reach customer DBs.

2. Workspace scanDetails — required for RAC clusters. Without it, the SCAN listener's redirect to a node-local IP (10.x.x.x) is unreachable from the executor → ORA-17820. Configure once via the AIDP workspace REST API or console:

   networkConfigurationDetails: {
     "subnetId": "ocid1.subnet.oc1.iad.aaaa...",
     "scanDetails": [
       {"fqdn": "<scan-host>.clientsubnet.dns.oraclevcn.com", "port": "1521"}
     ]
   }
   

   This activates PE-ARCH 3c (RCE with SCAN Proxy) — the PAGW translates RAC node-redirect IPs to Class-E NAT addresses so the JDBC redirect lands on the right tunnel.

Smoke-test connectivity from a notebook before chasing JDBC errors:

import socket, ipaddress
ip = socket.gethostbyname(SCAN_HOST)            # should be Class-E (255.x) or RFC-1918 — never public
with socket.create_connection((SCAN_HOST, 1521), timeout=15): pass

Read (inline connector options)

Use the official AIDP Exadata connector first:

import os
from oracle_ai_data_platform_connectors.aidataplatform import (
    AIDP_FORMAT, aidataplatform_options,
)

opts = aidataplatform_options(
    type="ORACLE_EXADATA",
    host=os.environ["EXACS_HOST"],
    port=int(os.environ.get("EXACS_PORT", "1521")),
    database_name=os.environ["EXACS_DATABASE_NAME"],
    user=os.environ["EXACS_USER"],
    password=os.environ["EXACS_PASSWORD"],
    schema=os.environ["EXACS_SCHEMA"],
    table=os.environ["EXACS_TABLE"],
)
df = spark.read.format(AIDP_FORMAT).options(**opts).load()
df.show()

Write

opts["table"] = os.environ["EXACS_TARGET_TABLE"]
opts["write.mode"] = os.environ.get("EXACS_WRITE_MODE", "APPEND")  # CREATE | APPEND | OVERWRITE | MERGE
df.write.format(AIDP_FORMAT).options(**opts).save()

Use catalog.id

catalog_opts = aidataplatform_options(
    type="ORACLE_EXADATA",
    schema=os.environ["EXACS_SCHEMA"],
    table=os.environ["EXACS_TABLE"],
    extra={"catalog.id": os.environ["EXACS_CATALOG_ID"]},
)
df = spark.read.format(AIDP_FORMAT).options(**catalog_opts).load()

Pushdown

pushdown_opts = aidataplatform_options(
    type="ORACLE_EXADATA",
    host=os.environ["EXACS_HOST"],
    port=int(os.environ.get("EXACS_PORT", "1521")),
    database_name=os.environ["EXACS_DATABASE_NAME"],
    user=os.environ["EXACS_USER"],
    password=os.environ["EXACS_PASSWORD"],
    extra={
        "pushdown.sql": "SELECT * FROM HR.EMPLOYEES WHERE DEPARTMENT_ID = 10",
    },
)
df = spark.read.format(AIDP_FORMAT).options(**pushdown_opts).load()

Gotchas

• scanDetails is the #1 cause of ORA-17820. If the JDBC URL points at the SCAN listener on a RAC cluster but the workspace doesn't have the SCAN FQDN registered, the redirect fails silently. See "Critical infrastructure prerequisites" above.
• Port 1521 is often encrypted by server-side NNE. The ORACLE_EXADATA connector is the preferred path; fall back to raw JDBC only for diagnostics that the built-in connector cannot express.
• No IMDS access from AIDP notebooks — Instance Principal / Resource Principal flows that work elsewhere on OCI compute fail here. Use API Key + inline PEM if you need OCI SDK calls.
• If you need wallet+TCPS or IAM DB-Token to an Autonomous DB, use the aidp-alh skill instead.

References

• Helper: scripts/oracle_ai_data_platform_connectors/aidataplatform.py
• Official Exadata sample: data-engineering/ingestion/Read_Write_Oracle_Ecosystem_Connectors/Oracle_Exadata.ipynb
• AIDP private endpoint design (PE-ARCH 1a–3c, SCAN Proxy): see workspace memory oci_private_endpoint_design.md.