Dockerfile.connect

FROM quay.io/strimzi/kafka:1.2.0-kafka-4.3.1 AS base
# ─── Stage 1: Download connector plugins ─────────────────────────────────────
#
# This stage downloads all enterprise connector plugins from Maven Central
# and GitHub. Each plugin is placed in its own directory under /plugins.
#
# Plugin versions are pinned as build args for easy upgrades.
#
FROM eclipse-temurin:21-jre-alpine AS download

ARG DEBEZIUM_VERSION=3.6.2.Final
ARG APICURIO_VERSION=3.3.0
ARG GROOVY_VERSION=5.0.7
ARG AIVEN_JDBC_VERSION=6.10.0
ARG AIVEN_S3_VERSION=3.4.3

RUN apk add --no-cache curl tar

WORKDIR /plugins

# ── Debezium CDC Connectors ──────────────────────────────────────────────────
# Debezium captures row-level changes from database transaction logs (WAL/binlog)
# and streams them as Kafka events. Essential for Change Data Capture pipelines.

# PostgreSQL — primary relational database CDC via logical replication
RUN mkdir -p debezium-postgres && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-connector-postgres/${DEBEZIUM_VERSION}/debezium-connector-postgres-${DEBEZIUM_VERSION}-plugin.tar.gz" \
    | tar -xz -C debezium-postgres --strip-components=1

# MySQL / MariaDB — binlog-based CDC for legacy OLTP systems
RUN mkdir -p debezium-mysql && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-connector-mysql/${DEBEZIUM_VERSION}/debezium-connector-mysql-${DEBEZIUM_VERSION}-plugin.tar.gz" \
    | tar -xz -C debezium-mysql --strip-components=1

# MongoDB — oplog/change-stream CDC for document stores
RUN mkdir -p debezium-mongodb && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-connector-mongodb/${DEBEZIUM_VERSION}/debezium-connector-mongodb-${DEBEZIUM_VERSION}-plugin.tar.gz" \
    | tar -xz -C debezium-mongodb --strip-components=1

# SQL Server — CDC via SQL Server Change Tracking
RUN mkdir -p debezium-sqlserver && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-connector-sqlserver/${DEBEZIUM_VERSION}/debezium-connector-sqlserver-${DEBEZIUM_VERSION}-plugin.tar.gz" \
    | tar -xz -C debezium-sqlserver --strip-components=1

# JDBC Sink Connector — Write events to relational databases
#
# The plugin archive bundles ojdbc11 (Oracle's driver, binary-only under Oracle's
# free-use terms — free to redistribute, but not open source). It is deleted so
# this image stays entirely open source; see the matching note on aiven-jdbc.
# Drivers for PostgreSQL, MySQL, MariaDB and SQL Server stay.
RUN mkdir -p debezium-jdbc && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-connector-jdbc/${DEBEZIUM_VERSION}/debezium-connector-jdbc-${DEBEZIUM_VERSION}-plugin.tar.gz" \
    | tar -xz -C debezium-jdbc --strip-components=1 && \
    rm -f debezium-jdbc/ojdbc*.jar

# ── Aiven JDBC Source/Sink Connector ─────────────────────────────────────────
# Generic JDBC source (io.aiven.connect.jdbc.JdbcSourceConnector) that polls
# relational tables into Kafka topics, plus a matching sink. Not published to
# Maven Central — Aiven attaches release archives to GitHub releases only
# (plain .tar, not gzipped).
#
# The archive bundles ojdbc8 alongside the open-source drivers. Oracle's JDBC
# driver ships as a binary under the Oracle Free Use Terms — free of charge and
# free to redistribute, but with no source and no OSI licence, so it is the one
# thing that would make this image not open source. It is removed here; Oracle
# is not a supported target for this image. Anyone who needs it can layer the
# driver back in:
#
#     FROM ghcr.io/bmscomp/connect:3.6.2
#     ARG OJDBC=https://repo1.maven.org/maven2/com/oracle/database/jdbc/ojdbc11/23.26.1.0.0/ojdbc11-23.26.1.0.0.jar
#     ADD --chown=1001:1001 ${OJDBC} /opt/kafka/plugins/aiven-jdbc/
#
# The connectors' own Oracle dialect classes are harmless without a driver: they
# are selected from the JDBC URL, so an Oracle URL now fails at connection time
# with "No suitable driver" instead of half-working.
RUN mkdir -p aiven-jdbc && \
    curl -sfL "https://github.com/Aiven-Open/jdbc-connector-for-apache-kafka/releases/download/v${AIVEN_JDBC_VERSION}/jdbc-connector-for-apache-kafka-${AIVEN_JDBC_VERSION}.tar" \
    | tar -x -C aiven-jdbc --strip-components=1 && \
    rm -f aiven-jdbc/ojdbc*.jar

# ── Aiven S3 Sink / Source Connectors ────────────────────────────────────────
# Archive Kafka topics to Amazon S3 and replay S3 objects back into Kafka.
# Apache-2.0 throughout — the connectors themselves, the AWS SDK for Java v2
# they bundle, and the rest of the archive. Confluent's kafka-connect-s3 is
# source-available under the Confluent Community Licence instead: redistribution
# is allowed, but it carries a competing-hosted-service restriction that this
# image would pass on to everyone who pulls it. Same release-archive shape as
# the JDBC connector above: plain .tar, one top-level directory.
#
# Sink and source ship as separate archives that each bundle their own copy of
# the shared commons jars, so they get separate plugin directories. Putting
# them in one directory would leave duplicate classes on the same plugin path.
RUN mkdir -p aiven-s3-sink && \
    curl -sfL "https://github.com/Aiven-Open/cloud-storage-connectors-for-apache-kafka/releases/download/v${AIVEN_S3_VERSION}/s3-sink-connector-for-apache-kafka-${AIVEN_S3_VERSION}.tar" \
    | tar -x -C aiven-s3-sink --strip-components=1

# The source release archive also ships its dependencies *exploded*: about 34k
# loose .class files and resources sitting next to the jars. Every one of them
# already exists inside a bundled jar — verified entry by entry against the
# 3.4.3 archive, zero unique paths — so they are pure duplication. Dropping them
# takes the plugin directory from ~300 MB to ~65 MB and keeps a second,
# unversioned copy of Avro, Hadoop and the AWS SDK off the plugin classpath.
# Licence and notice texts are kept.
RUN mkdir -p aiven-s3-source && \
    curl -sfL "https://github.com/Aiven-Open/cloud-storage-connectors-for-apache-kafka/releases/download/v${AIVEN_S3_VERSION}/s3-source-connector-for-apache-kafka-${AIVEN_S3_VERSION}.tar" \
    | tar -x -C aiven-s3-source --strip-components=1 && \
    cd aiven-s3-source && \
    for entry in *; do \
        case "$entry" in \
            *.jar|LICENSE*|NOTICE*|licenses|notices) ;; \
            *) rm -rf "$entry" ;; \
        esac; \
    done

# Avro and Parquet in both S3 archives carry CRITICAL CVEs fixed in patch
# releases: CVE-2024-47561 (Avro schema parsing, RCE; sink bundles 1.11.0,
# source 1.11.3) and CVE-2025-30065 (parquet-avro 1.15.0, arbitrary code
# execution). 3.4.3 is Aiven's latest, so each plugin's copies are swapped for
# the newest patch of the same line: Avro 1.11.5, and every parquet-* jar at
# 1.15.2 so the Parquet modules stay in step (1.11.5 and 1.15.2 also fix
# CVE-2025-33042 and CVE-2025-46762). The build fails if an archive stops
# bundling the jar it replaces; that is the cue to drop the override.
ARG AVRO_VERSION=1.11.5
ARG PARQUET_VERSION=1.15.2
RUN for dir in aiven-s3-sink aiven-s3-source; do \
        old=$(ls "$dir"/avro-1.11.[0-4].jar) && rm "$old" && \
        curl -sfL "https://repo1.maven.org/maven2/org/apache/avro/avro/${AVRO_VERSION}/avro-${AVRO_VERSION}.jar" \
            -o "$dir/avro-${AVRO_VERSION}.jar" || exit 1; \
        for m in parquet-avro parquet-column parquet-common parquet-encoding \
                 parquet-format-structures parquet-hadoop parquet-jackson; do \
            rm "$dir/$m-1.15.0.jar" && \
            curl -sfL "https://repo1.maven.org/maven2/org/apache/parquet/$m/${PARQUET_VERSION}/$m-${PARQUET_VERSION}.jar" \
                -o "$dir/$m-${PARQUET_VERSION}.jar" || exit 1; \
        done; \
    done

# ── Apicurio Schema Registry Integration ─────────────────────────────────────
# Apicurio converters enable schema governance with Avro, JSON Schema, and
# Protobuf serialization. Required when connecting to Apicurio Registry.
#
# Use Apicurio's Connect-converter distribution, not the bare serde jars: it
# ships the converters together with every dependency they import (the Apicurio
# Java SDK, Avro, Protobuf, Jackson, netty). Downloading the four serde jars on
# their own left every import unresolved, so Connect logged a LinkageError per
# converter during the plugin scan and started without any of them — a failure
# invisible to the build. The archive is flat, so no --strip-components.
RUN mkdir -p apicurio-converter && \
    curl -sfL "https://repo1.maven.org/maven2/io/apicurio/apicurio-registry-distro-connect-converter/${APICURIO_VERSION}/apicurio-registry-distro-connect-converter-${APICURIO_VERSION}.tar.gz" \
    | tar -xz -C apicurio-converter

# Debezium Scripting SMT and Groovy JSR-223 Engine
#
# The scripting archive holds the SMT and nothing else. Alone on its own plugin
# path it cannot resolve the Debezium runtime classes it compiles against
# (io.debezium.config.EnumeratedValue, io.debezium.metadata.ConfigDescriptor,
# ...), so Connect dropped both transformations during the plugin scan and
# started without them. Debezium 3.6 splits that runtime across small modules —
# there is no debezium-core artifact any more — so pull the same four a
# connector plugin directory carries.
RUN mkdir -p debezium-scripting && \
    curl -sfL "https://repo1.maven.org/maven2/io/debezium/debezium-scripting/${DEBEZIUM_VERSION}/debezium-scripting-${DEBEZIUM_VERSION}.tar.gz" \
    | tar -xz -C debezium-scripting --strip-components=1 && \
    for jar in groovy groovy-jsr223 groovy-json; do \
        curl -sfL "https://repo1.maven.org/maven2/org/apache/groovy/${jar}/${GROOVY_VERSION}/${jar}-${GROOVY_VERSION}.jar" \
        -o "debezium-scripting/${jar}-${GROOVY_VERSION}.jar"; \
    done && \
    for jar in debezium-api debezium-config debezium-util debezium-connector-common; do \
        curl -sfL "https://repo1.maven.org/maven2/io/debezium/${jar}/${DEBEZIUM_VERSION}/${jar}-${DEBEZIUM_VERSION}.jar" \
        -o "debezium-scripting/${jar}-${DEBEZIUM_VERSION}.jar"; \
    done

# ─── Stage 2: Final production image ─────────────────────────────────────────
FROM base

ARG DEBEZIUM_VERSION=3.6.2.Final
ARG AIVEN_JDBC_VERSION=6.10.0
ARG AIVEN_S3_VERSION=3.4.3

USER root:root
COPY --from=download /plugins/ /opt/kafka/plugins/
# The base image is Strimzi's Kafka image, which also carries the Kafka
# Exporter for the operator's exporter Deployment. That Deployment runs
# Strimzi's own image, never this one, and a Connect worker never starts the
# exporter; its binary is built with Go 1.24.0 (CVE-2025-68121, CRITICAL,
# crypto/tls). Removing it takes the CVE out of the image.
RUN rm -rf /opt/kafka-exporter
USER 1001

# ─── OCI Image Labels (https://github.com/opencontainers/image-spec) ─────────
LABEL org.opencontainers.image.title="connect"
LABEL org.opencontainers.image.description="Kafka Connect image with fully open-source Debezium CDC (PostgreSQL, MySQL, MongoDB, SQL Server), Apicurio Schema Registry (Avro, JSON Schema, Protobuf), Debezium JDBC Sink, Aiven JDBC source/sink, and Aiven S3 sink/source. Built on Strimzi Kafka 4.3.1."
LABEL org.opencontainers.image.source="https://github.com/bmscomp/kates"
LABEL org.opencontainers.image.url="https://github.com/bmscomp/kates/pkgs/container/connect"
LABEL org.opencontainers.image.documentation="https://github.com/bmscomp/kates/blob/main/docs/kafka-connect.md"
LABEL org.opencontainers.image.vendor="bmscomp"
# Every bundled jar is open source, but not all of it is Apache-2.0: the JDBC
# drivers and a few transitive dependencies bring MIT, BSD, LGPL, CDDL/EPL and
# (for MySQL Connector/J) GPLv2-with-FOSS-exception terms. The OCI spec wants the
# licences of the software actually contained in the image, so list them all.
LABEL org.opencontainers.image.licenses="Apache-2.0 AND MIT AND BSD-2-Clause AND BSD-3-Clause AND LGPL-2.1-or-later AND CDDL-1.0 AND CDDL-1.1 AND EPL-2.0 AND (GPL-2.0-only WITH Universal-FOSS-exception-1.0)"
LABEL org.opencontainers.image.base.name="quay.io/strimzi/kafka:1.2.0-kafka-4.3.1"
LABEL io.debezium.version="${DEBEZIUM_VERSION}"
LABEL io.aiven.jdbc.version="${AIVEN_JDBC_VERSION}"
LABEL io.aiven.s3.version="${AIVEN_S3_VERSION}"