Doris compose

Use doris compose to create doris docker compose clusters.

Requirements

1. Make sure you have docker permissions

run:

docker run hello-world

if have problem with permission denied, then add-docker-permission.

Make sure BuildKit configured in the machine. if not follow docker-with-BuildKit.

2. The doris image should contains

/opt/apache-doris/{fe, be, ms}

If don't create cloud cluster, the image no need to contains the ms pkg.

If build doris use sh build.sh --fe --be --cloud without do any change on their origin conf or shells, then its output/ satisfy with all above, then run command in doris root directory will generate such a image. If you want to pack a product that is not the output/ directory, you can modify Dockerfile by yourself.

docker build -f docker/runtime/doris-compose/Dockerfile -t <image> .

The Dockerfile default use JDK 17, for doris 2.1, 3.0, master, they all default use JDK 17.

But doris 2.0 still use JDK 8, for build 2.0 image, user need specific use JDK 8 with arg JDK_IMAGE=openjdk:8u342-jdk. Here is build 2.0 image command:

docker build -f docker/runtime/doris-compose/Dockerfile \
     --build-arg JDK_IMAGE=openjdk:8u342-jdk            \
    -t <image> .

The <image> is the name you want the docker image to have.

User can also download a doris release package from Doris Home or Doris Github, extract it, then build its image with arg OUTPUT_PATH

for example:

cd ~/tmp
wget https://apache-doris-releases.oss-accelerate.aliyuncs.com/apache-doris-3.0.5-bin-x64.tar.gz
tar xvf apache-doris-3.0.5-bin-x64.tar.gz  # after extract, there will be a directory ./apache-doris-3.0.5-bin-x64/{fe, be, ms}

# -f: the Dockerfile file
# -t: the builded image
# . : current directory, here it's ~/tmp, then output path is ~/tmp/apache-doris-3.0.5-bin-x64
docker build \
    --build-arg OUTPUT_PATH=./apache-doris-3.0.5-bin-x64 \
    -f ~/workspace/doris/docker/runtime/doris-compose/Dockerfile \
    -t my-doris:v3.0.5 \
    .

3. Install the dependent python library in ‘docker/runtime/doris-compose/requirements.txt’

PyYAML of certain version not always fit other libraries' requirements. So we suggest to use a individual environment using venv or conda.

python -m pip install --user -r docker/runtime/doris-compose/requirements.txt

if it failed, change content of requirements.txt to:

pyyaml==5.3.1
docker==6.1.3
......

Usage

Notice

Each cluster will have a directory in ‘/tmp/doris/{cluster-name}’, user can set env LOCAL_DORIS_PATH to change its directory.

For example, if user export LOCAL_DORIS_PATH=/mydoris, then the cluster's directory is ‘/mydoris/{cluster-name}’.

And cluster‘s directory will contains all its containers’s logs and data, like fe-1, fe-2, be-1, ..., etc.

If there are multiple users run doris-compose on the same machine, suggest don't change LOCAL_DORIS_PATH or they should export the same LOCAL_DORIS_PATH.

Because when create a new cluster, doris-compose will search the local doris path, and choose a docker network which is different with this path's clusters.

So if multiple users use different LOCAL_DORIS_PATH, their clusters may have docker network conflict!!!

Create a cluster or recreate its containers

python docker/runtime/doris-compose/doris-compose.py up  <cluster-name>   <image?>
    --add-fe-num  <add-fe-num>  --add-be-num <add-be-num>
    [--fe-id <fd-id> --be-id <be-id>]
    ...
    [ --cloud ]
    [ --cluster-snapshot <cluster-snapshot-json> ]

if it's a new cluster, must specific the image.

add fe/be nodes with the specific image, or update existing nodes with --fe-id, --be-id

The --cluster-snapshot parameter allows you to provide a cluster snapshot JSON content for FE-1 first startup in cloud mode only. The JSON will be written to FE conf/cluster_snapshot.json and passed to start_fe.sh with --cluster_snapshot parameter. This is only effective on first startup.

Example:

python docker/runtime/doris-compose/doris-compose.py up my-cluster my-image --cloud --cluster-snapshot '{"instance_id":"instance_id_xxx"}'

For create a cloud cluster, steps are as below:

1. Write cloud s3 store config file, its default path is ‘/tmp/doris/cloud.ini’. It's defined in environment variable DORIS_CLOUD_CFG_FILE, user can change this env var to change its path. A Example file is locate in ‘docker/runtime/doris-compose/resource/cloud.ini.example’.

2. Use doris compose up command with option --cloud to create a new cloud cluster.

The simplest way to create a cloud cluster:

python docker/runtime/doris-compose/doris-compose.py up  <cluster-name>  <image>  --cloud

To create a cloud cluster with a custom cluster snapshot:

python docker/runtime/doris-compose/doris-compose.py up  <cluster-name>  <image>  --cloud --cluster-snapshot '{"instance_id":"instance_id_xxx"}'

It will create 1 fdb, 1 meta service server, 1 recycler, 3 fe and 3 be.

Remove node from the cluster

python docker/runtime/doris-compose/doris-compose.py down  <cluster-name> --fe-id <fe-id>  --be-id<be-id> [--clean]  [--drop-force]

Down the containers and remove it from the DB.

For BE, if specific drop force, it will send dropp sql to FE, otherwise it will send decommission sql to FE.

If specific --clean, it will delete its data too.

Start, stop, restart specific nodes

python docker/runtime/doris-compose/doris-compose.py start  <cluster-name>  --fe-id  <multiple fe ids>  --be-id <multiple be ids>
python docker/runtime/doris-compose/doris-compose.py restart  <cluster-name>  --fe-id  <multiple fe ids>  --be-id <multiple be ids>

List doris cluster

python docker/runtime/doris-compose/doris-compose.py ls <multiple cluster names>

if specific cluster names, it will list all the cluster's nodes.

Otherwise it will just list summary of each clusters.

There are more options about doris-compose. Just try

python docker/runtime/doris-compose/doris-compose.py <command> -h

Docker suite in regression test

Regression test support running a suite in a docker doris cluster.

See the example demo_p0/docker_action.groovy.

The docker suite can specify fe num and be num, and add/drop/start/stop/restart the fe and be.

Before run a docker suite, read the annotation in demo_p0/docker_action.groovy carefully.

Generate regression custom conf file

provide a command for let the regression test connect to a docker cluster.

python docker/runtime/doris-compose/doris-compose.py config <cluster-name>  <doris-root-path>  [-q]  [--connect-follow-fe]

Generate regression-conf-custom.groovy to connect to the specific docker cluster.

Setup cloud multi clusters test env

steps:

1. Create a new cluster: python docker/runtime/doris-compose/doris-compose.py up my-cluster my-image --add-fe-num 2 --add-be-num 4 --cloud
2. Generate regression-conf-custom.groovy: python docker/runtime/doris-compose/doris-compose.py config my-cluster <doris-root-path> --connect-follow-fe
3. Run regression test: bash run-regression-test.sh --run -times 1 -parallel 1 -suiteParallel 1 -d cloud/multi_cluster

Multi cloud cluster with shared Meta Service

Doris compose now supports creating multiple cloud clusters that share the same Meta Service (MS), FDB, and Recycler services. This is useful for testing cross-cluster operations (such as cloning, backup/restore) under the same Meta Service instance.

Create the first cluster

First, create a complete cloud cluster that will provide MS/FDB/Recycler services:

python docker/runtime/doris-compose/doris-compose.py up cluster1 <image> --cloud --add-fe-num 1 --add-be-num 3

This creates the first cluster with:

• 1 FDB node
• 1 Meta Service (MS) node
• 1 Recycler node
• 1 FE node
• 3 BE nodes

Create additional clusters sharing the same MS

Now you can create additional sql/compute clusters that share the first cluster's Meta Service:

# Create second cluster sharing cluster1's MS
python docker/runtime/doris-compose/doris-compose.py up cluster2 <image> --cloud --external-ms cluster1 --instance-id instance_cluster2 --add-fe-num 1 --add-be-num 3

# Create third cluster sharing cluster1's MS
python docker/runtime/doris-compose/doris-compose.py up cluster3 <image> --cloud --external-ms cluster1 --instance-id instance_cluster3 --add-fe-num 1 --add-be-num 3

Key points:

• --external-ms cluster1: Specifies that this cluster will use cluster1's MS/FDB/Recycler services
• --instance-id: Must be unique for each cluster. If not specified, will auto-generate as instance_<cluster-name>
• The new clusters will NOT create their own MS/FDB/Recycler nodes, saving resources
• All clusters share the same object storage and meta service infrastructure
• Each cluster maintains its own FE/BE nodes for compute isolation

Network architecture

When using external MS:

• Each cluster has its own Docker network
• Compute clusters join the external MS cluster's network as well
• DNS resolution is configured automatically for all MS/FDB/Recycler nodes
• BE and FE nodes can communicate with MS nodes using their container names

Validation

Doris compose automatically validates:

1. External MS cluster exists
2. External cluster is a cloud cluster
3. MS and FDB nodes are present
4. MS and FDB containers are running

If validation fails, you'll get a clear error message explaining what needs to be fixed.

Rollback Cloud Cluster to Snapshot

The rollback command allows you to rollback a cloud cluster to a specific snapshot state.

Basic Usage

python docker/runtime/doris-compose/doris-compose.py rollback <cluster-name> \
    --cluster-snapshot '{"instance_id":"instance_xxx", ...}' \
    [--instance-id NEW_INSTANCE_ID]

What it does

The rollback command performs the following operations on ALL FE/BE nodes:

1. Stops all FE and BE nodes
2. Cleans FE doris-meta/ and BE storage/ directories (preserves conf/, log/, etc.)
3. Updates update all nodes conf
4. Restarts all nodes with new instance_id and cluster_snapshot

Parameters

• --cluster-snapshot (required): Cluster snapshot JSON content

  • Example: '{"instance_id":"instance_id_xxx"}'
  • Will be written to FE-1's conf/cluster_snapshot.json

• --instance-id (optional): New instance ID after rollback

  • If not specified, auto-generates: instance_{cluster_name}_{timestamp}

• --wait-timeout (optional): Wait seconds for nodes to be ready (default: 0)

Examples

Full cluster rollback:

python docker/runtime/doris-compose/doris-compose.py rollback my_cluster \
    --cluster-snapshot '{"instance_id":"backup_instance", ...}' \
    --wait-timeout 60

Rollback with custom instance ID:

python docker/runtime/doris-compose/doris-compose.py rollback my_cluster \
    --cluster-snapshot '{"instance_id":"rollback_instance", ...}' \
    --instance-id "prod_rollback_20251027"

Problem investigation

Log

Each cluster has logs in Docker in ‘/tmp/doris/{cluster-name}/{node-xxx}/log/’. For each node, doris compose will also print log in ‘/tmp/doris/{cluster-name}/{node-xxx}/log/health.out’

Core Dump

Doris Compose supports core dump generation for debugging purposes. When a process crashes, it will generate a core dump file that can be analyzed with tools like gdb.

Core Dump Location

Core dump files are generated in the following locations:

• Host System: /tmp/doris/{cluster-name}/{node-xxx}/core_dump/
• Container: /opt/apache-doris/core_dump/

The core dump files follow the pattern: core.{executable}.{pid}.{timestamp}

For example:

/tmp/doris/my-cluster/be-1/core_dump/core.doris_be.12345.1755418335

Core Pattern Configuration

The system uses the core pattern from /proc/sys/kernel/core_pattern on the host system. The default pattern is:

/opt/apache-doris/core_dump/core.%e.%p.%t

Where:

• %e: executable name
• %p: process ID
• %t: timestamp

Core Dump Settings

Doris Compose automatically configures the following settings for core dump generation:

1. Container Settings:

   • ulimits.core = -1 (unlimited core file size)
   • cap_add: ["SYS_ADMIN"] (required capabilities)
   • privileged: true (privileged mode)

2. Directory Permissions:

   • Core dump directory is created with 777 permissions
   • Ownership is set to the host user for non-root containers

3. Non-Root User Support:

   • Core dump directory permissions are automatically configured
   • Works with both root and non-root user containers

Troubleshooting

If core dumps are not being generated:

1. Check ulimit settings:

   ulimit -c
   # Should return "unlimited" or a positive number
   

2. Check directory permissions:

   ls -la /tmp/doris/{cluster-name}/{node-xxx}/core_dump/
   # Should show 777 permissions
   

3. Check core pattern:

   cat /proc/sys/kernel/core_pattern
   # Should show the expected pattern
   

4. Check container logs:

   docker logs {container-name}
   # Look for core dump related messages
   

Up cluster using non-detach mode

python docker/runtime/doris-compose/doris-compose.py up ...   -no-detach

Developer

Before submitting code, pls format code.

bash format-code.sh