Manage transfer agents

Storage Transfer Service agents are applications running inside a Docker container, that coordinate with Storage Transfer Service to read data from POSIX file system sources, and/or write data to POSIX file system sinks.

If your transfer does not involve a POSIX file system, you do not need to set up agents.

This document describes how to administer transfer agents on your servers.

Overview

• Agent processes are dynamic. While you are running a transfer, you can add agents to increase performance. Newly started agents join the assigned agent pool and perform work from existing transfers. You can use this to adjust how many agents are running, or to adapt transfer performance to changing transfer demand.

• Agent processes are a fault-tolerant collective. If one agent stops running, the remaining agents continue to do work. If all of your agents stop, when you restart the agents the transfer resumes where the agents stopped. This enables you to avoid monitoring agents, retrying transfers, or implementing recovery logic. You can patch, move, and dynamically scale your agent pools without transfer downtime by coordinating agents with Google Kubernetes Engine.

  For example, you submit two transfers while two agents are running. If one of the agents stops due to a machine reboot or operating system patch, the remaining agent continues working. The two transfers are still running, but slower since a single agent is moving data. If the remaining agent also stops, then all transfers stop making progress, since there are no agents running. When you restart the agent processes, the transfers resume where they left off.

• Agent processes belong to a pool. They collectively move your data in parallel. Because of this, all agents within a pool must have the same access to all data sources that you want to transfer.

  For example, if you are transferring data from a particular file system, you must mount the file system to every machine that is hosting agents in your agent pool. If some agents in your pool can reach a data source and others can't, transfers from that data source won't succeed.

Before you begin

Before configuring your transfers, make sure you have configured access: for users and service accounts.

If you'll be using gcloud commands, install the gcloud CLI.

Install and run transfer agents

Important: Users able to create transfer jobs can retrieve data from, and download data to, any file system directory that is accessible by the agent. If agents are run as root and are given access to the entire file system, a malicious actor may be able to take over the host. It is strongly recommended that you restrict agent access to only necessary directories.

We recommend installing a minimum of three agents per agent pool, ideally on separate machines. For more information about determining how many agents to run, see Maximizing transfer agent performance.

To install and run transfer agents:

April 12, 2023: Agents being used for transfers from S3-compatible storage must authenticate with a service account key. Use the --creds-file flag to specify the path to a service account key file. Failure to use a service account results in stalled transfers; the job shows as in progress but no data is transferred.

This notice will be removed when the issue is resolved.

Google Cloud console

1. In the Google Cloud console, go to the Agent pools page.

   Go to Agent pools

2. Select the agent pool to which to add the new agent.

3. Click Install agent.

4. Follow the instructions to install and run the agent.

   For more information about the agent's command-line options, see Agent command-line options.

gcloud CLI

To install one or more agents using the gcloud CLI, run gcloud transfer agents install:

gcloud transfer agents install --pool=POOL_NAME --count=NUM_AGENTS \
  --mount-directories=MOUNT_DIRECTORIES

The tool walks you through any required steps to install the agent(s). This command installs NUM_AGENTS agent(s) on your machine, mapped to the pool name specified as POOL_NAME, and authenticates the agent using your gcloud credentials. The pool name must exist, or an error is returned.

The --mount-directories flag is optional but is strongly recommended. Its value is a comma-separated list of directories on the file system to which to grant the agent access. Omitting this flag mounts the entire file system to the agent container. See the gcloud reference for more details.

S3-compatible sources

When installing agents to use with an S3-compatible source, you must provide AWS credentials either as environment variables as the values of AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY, or stored as default credentials in your system's configuration files.

export AWS_ACCESS_KEY_ID=ID
export AWS_SECRET_ACCESS_KEY=SECRET
gcloud transfer agents install --pool=POOL_NAME \
  --creds-file=/relative/path/to/service-account-key.json

Use a service account key

To run agents using a service account key, use the --creds-file option:

gcloud transfer agents install --pool=POOL_NAME --count=NUM_AGENTS \
   --creds-file=/relative/path/to/service-account-key.json

More info

For a full list of optional flags, run gcloud transfer agents install --help or read the gcloud transfer reference.

docker run

Before using docker run to install agents, follow the instructions to install Docker.

The docker run command installs one agent. To increase the number of agents in your pool, re-run this command as many times as required.

When installing agents, you can choose to authenticate using your gcloud default credentials, or with a service account.

Default credentials

To allow the Docker container to authenticate with your gcloud default credentials, create a Docker volume containing a file with your application default credentials by running the following command:

sudo docker run -ti --name gcloud-config google/cloud-sdk gcloud auth application-default login

Then use the following command to install an agent, using the --volumes-from flag to mount the gcloud-config credentials volume:

sudo docker run --ulimit memlock=64000000 -d --rm \
--volumes-from gcloud-config \
-v HOST_DIRECTORY:CONTAINER_DIRECTORY \
gcr.io/cloud-ingest/tsop-agent:latest \
--project-id=PROJECT_ID \
--hostname=$(hostname) \
--agent-pool=POOL_NAME

Service account auth

To install and run transfer agents docker run using service account credentials, specify the path to your JSON-formatted service account key using the --creds-file flag.

The path must be prefixed with the string, /transfer_root.

See Create and manage service account keys for more information about service account keys.

sudo docker run --ulimit memlock=64000000 -d --rm \
-v HOST_DIRECTORY:CONTAINER_DIRECTORY \
-v PATH/TO/KEY.JSON:PATH/TO/KEY.JSON \
gcr.io/cloud-ingest/tsop-agent:latest \
--project-id=PROJECT_ID \
--creds-file=/transfer_root/PATH/TO/KEY.JSON
--hostname=$(hostname) \
--agent-pool=POOL_NAME

Options and flags

Replace the variables in the examples above with the following information:

• HOST_DIRECTORY is the directory on the host machine that you intend to copy from. You can use more than one -v flag to specify additional directories to copy from.
• CONTAINER_DIRECTORY is the directory mapped within the agent container. It must be the same as HOST_DIRECTORY.
• PROJECT_ID is the project ID that is hosting the transfer.
• POOL_NAME is the name of the agent pool in which to install this agent. If you omit this flag, the agent is installed into your project's transfer_service_default pool.

The docker run command supports additional flags.

• --enable-mount-directory mounts the entire file system under the /transfer_root directory on the container. If --enable-mount-directory is specified, directory restrictions using the -v flag are not applied.

  Caution: If agents are run as root and are given access to the entire file system, a malicious actor may be able to take over the host. It is strongly recommended that you restrict agent access to only necessary directories.

• --creds-file=CREDENTIAL_FILE specifies the path to a JSON-formatted service account credential file. Unless you're using --enable_mount_directory, you must:

  1. Mount the credential file using the -v flag.
  2. Prefix the path to --creds-file with /transfer_root.

  For example:

  -v /tmp/key.json:/tmp/key.json \
  --creds-file=/transfer_root/tmp/key.json
  

• --enable-s3 specifies that this agent is for transfers from S3-compatible storage. Agents installed with this option cannot be used for transfers from POSIX file systems.

• If your transfer is from AWS S3 or S3-compatible storage, pass your access key ID and secret key using environment variables:

  -e AWS_ACCESS_KEY_ID=AWS_ACCESS_KEY_ID \
  -e AWS_SECRET_ACCESS_KEY=AWS_SECRET_ACCESS_KEY
  

• --env HTTPS_PROXY=PROXY specifies a forward proxy on your network. The value of PROXY is the HTTP URL and port of the proxy server. Ensure that you specify the HTTP URL, and not an HTTPS URL, to avoid double-wrapping requests in TLS encryption. Double-wrapped requests prevent the proxy server from sending valid outbound requests.

• --agent-id-prefix=ID_PREFIX specifies an optional prefix that is prepended to the agent ID to help identify the agent or its machine in the Google Cloud console. When a prefix is used, the agent ID is formatted as prefix + hostname + Docker container ID.

• --log-dir=LOGS_DIRECTORY modifies the directory to which the agent writes logs. The default directory is /tmp/.

  If you haven't specified --enable_mount_directory, you must prefix this path with /transfer_root. For example, /transfer_root/logs.

• --max-physical-mem=MAX_MEMORY: agents default to using 8GiB maximum of system memory. If the default doesn't fit your environment, you can specify a relevant maximum memory usage in the following formats:

  max-physical-mem value	Maximum memory setting
  6g	6 gigabytes
  6gb	6 gigabytes
  6GiB	6 gibibytes

Confirm agent connections

If you created your agents using the gcloud CLI or docker run, the agents won't show as connected until you've created the required Pub/Sub resources. You can create these resources in one of two ways:

• Create a transfer with this agent pool. Storage Transfer Service creates the necessary Pub/Sub resources automatically.

• Follow the instructions to install an agent with the Cloud console. When you reach the Install agents screen, find the section that states If you haven't already done so, click to create the required Cloud Pub/Sub topics and subscriptions and click Create.

To confirm that your agents are connected:

1. In the Google Cloud console, go to the Agent pools page.

   Go to Agent pools

   Your agent pools are displayed, with the number of connected agents.

2. Select an agent pool to view details on connected agents.

If a new agent doesn't show up in the agent pool page within 10 minutes of its creation, see Agents are not connected.

Monitor agent activity

You can use Cloud Monitoring alerts to monitor agent activity.

Monitoring is available along project, agent_pool, and agent_id dimensions.

Using this monitoring data, you can set up alerts to notify you of potential issues with your transfer. To do so, create an alert on either of the following Google Cloud metrics:

Metric name	What it describes	Suggested uses
storagetransfer.googleapis.com/agent/transferred_bytes_count	Measures how quickly a specific agent is moving data across all jobs that it services at a point in time.	Alert for dips in performance.
storagetransfer.googleapis.com/agent/connected	A boolean that is True for each agent that Google Cloud received a recent heartbeat message from.	Alert for failing agents Failing below a number of agents that you consider necessary for reasonable performance Signal an issue with agent machines

Stop an agent

To stop an agent, run docker stop on the agent's Docker container ID. To find the ID and stop the agent:

1. In the Google Cloud console, go to the Agent pools page.

   Go to Agent pools

2. Select the agent pool containing the agent to stop.

3. Select an agent from the list. Use the Filter field to search for prefixes, agent status, agent age, and more.

4. Click Stop agent. The docker stop command with the specific container ID is displayed.

5. Run the command on the machine on which the agent is running. A successful docker stop command returns the container ID.

Once stopped, the agent is shown in the agent pools list as Disconnected.

Delete an agent

To delete specific agents, list which agents are running on your machine:

docker container list --all --filter ancestor=gcr.io/cloud-ingest/tsop-agent

Then pass the agent IDs to transfer agents delete:

gcloud transfer agents delete --ids=id1,id2,…

To delete all agents running on the machine, use either the --all flag or the --uninstall flag. Both flags delete all agents on the machine; the --uninstall flag additionally uninstalls the agent Docker image.

gcloud transfer agents delete --all
gcloud transfer agents delete --uninstall

File system transfer details

Incremental transfers

Storage Transfer Service begins all transfers by computing the data present at the source and destination to determine which source files are new, updated, or deleted since the last transfer. We do this to reduce the amount of data we send from your machines, to use bandwidth effectively, and to reduce transfer times.

To detect whether a file has changed, we use an algorithm similar to gsutil rsync: we check the last modified time and size of the source file, and compare that to the last modified time and size recorded when the file was last copied. When we detect a new or changed file, we copy the entire file to its destination. For more information about file freshness, see Data consistency details.

By default we detect, but do not act on, files deleted on the source. If you choose the sync option Delete destination files that aren't also in the source when creating or editing, your transfer will delete the corresponding object at the destination.

If you choose the sync option Delete destination files that aren't also in the source, files that are accidentally deleted at the source are also deleted at the destination. To prevent data loss from accidental deletions, we recommend enabling object versioning in your destination bucket if you choose to use this option. Then, if you delete a file accidentally, you can restore your objects in Cloud Storage to an older version.

Data consistency details

A successful transfer operation will transfer all source files which existed and were not modified during the operation's entire running time. Source files that were created, updated, or deleted during a transfer may or may not have those changes reflected in the destination data set.

Storage Transfer Service uses a file's last modification time and size to determine if it changed. If a file is updated without changing its last modification time or size, and you enable the delete-objects-from-source option, you may lose data from that change.

When using the delete-objects-from-source feature, we strongly recommend that you freeze writes to the source for the duration of the transfer to protect against data loss.

To freeze writes to your source, do either of the following:

• Clone the directory you intend to transfer, and then use the cloned directory as the transfer source.
• Halt applications that write to the source directory.

If it's important to capture changes that occurred during a transfer, you can either re-run the transfer, or set the source file system as read-only while the operation is running.

Since Cloud Storage doesn't have the notion of directories, empty source directories are not transferred.

Important: We strongly advise that you validate that a transfer was successful before deleting any source data. This service doesn't have the capability to restore data in Cloud Storage.