Code Synchronization (Dev Mode)

You can synchronize your code (and other files) to and from running containers using dev mode.

Dev mode works similarly to the older hot reloading functionality, but is much faster and more reliable. It also supports bidirectional syncing, which enables you to sync new/changed files from your containers to your local machine.

This new sync mode uses Mutagen under the hood. Garden automatically takes care of fetching Mutagen, so you don't need to install any dependencies yourself to make use of dev mode.

Dev mode sync is not affected by the usual includes/excludes (e.g. rules defined in .gardenignore files), which makes it more flexible than hot reloading.

Instead, exclusion rules for dev mode are configured explicitly on the provider level and for each individual sync you configure—more on that below.

For example, you can use it to sync your build/dist directory into your container while running local, incremental builds (without having to remove those directories from your ignorefiles).

Please make sure to specify any paths that should not be synced by setting the provider-level default excludes and/or the exclude field on each configured sync! Otherwise you may end up syncing large directories and even run into application errors.

Configuration

To configure a service for dev mode, add devMode to your module/service configuration to specify your sync targets:

Configuring dev mode for container modules

kind: Module
name: node-service
type: container
services:
  - name: node-service
    args: [npm, start]
    devMode:
      command: [npm, run, dev] # Overrides the container's default when the service is deployed in dev mode
      sync:
        # Source/target configuration for dev mode is the same as for hot reloading.
        - source: src
          target: /app/src
          # Make sure to specify any paths that should not be synced!
          exclude: [node_modules]
          mode: two-way
  ...

Configuring dev mode for kubernetes and helm modules

kind: Module
type: kubernetes # this example looks the same for helm modules (i.e. with `type: helm`)
name: node-service
# For `kubernetes` and `helm` modules, the `devMode` field is located at the top level.
devMode:
  command: [npm, run, dev]
  sync:
    - target: /app
    - source: /tmp/somedir
      target: /somedir
serviceResource:
  kind: Deployment
  name: node-service-deployment
  containerModule: node-service-image
  containerName: node-service
...

Deploying with dev mode

To deploy your services with dev mode enabled, you can use the deploy or dev commands:

# Deploy specific services in dev mode:
garden deploy --dev myservice
garden deploy --dev myservice,my-other-service

# Deploy all applicable services in dev mode:
garden deploy --dev=*

# The dev command deploys services in dev mode by default:
garden dev myservice

Once your services have been deployed, any changes you make that fall under one of the sync specs you've defined will be automatically synced between your local machine and the running service.

Once you quit/terminate the Garden command, all syncs established by the command will be stopped (but the services will still be left running).

Sync modes

Garden's dev mode supports several sync modes, each of which maps onto a Mutagen sync mode.

In brief: It's generally easiest to get started with the one-way or two-way sync modes, and then graduate to a more fine-grained setup based on one-way-replica and/or one-way-replica-reverse once you're ready to specify exactly which paths to sync and which files/directories to ignore from the sync.

one-way-safe (or alias one-way)

  • Syncs a local source path to a remote target path.

  • When there are conflicts, does not replace/delete files in the remote target path.

  • Simple to use, especially when there are files/directories inside the remote target that you don't want to override with the contents of the local source.

  • On the other hand, if your setup / usage pattern is such that conflicts do sometimes arise for the source/target pair in question, you may want to use one-way-replica instead.

one-way-replica

  • Syncs a local source path to a remote target path, such that target is always an exact mirror of source (with the exception of excluded paths).

  • When using this mode, there can be no conflicts—the contents of source always override the contents of target.

  • Since conflicts are impossible here, this mode tends to be a better / more reliable choice long-term than one-way/one-way-safe. However, you may need to configure more fine-grained/specific source/target pairs and their excludes such that you don't have problems with paths in the remote target being overwritten/deleted when they change in the local source.

one-way-reverse

  • Same as one-way, except the direction of the sync is reversed.

  • Syncs a remote target path to a local source path.

  • Has the same benefits and drawbacks as one-way: Simple to configure, but conflicts are possible.

one-way-replica-reverse

  • Same as one-way-replica, except the direction of the sync is reversed.

  • Syncs a remote target path to a local source path, such that source is always an exact mirror of target (with the exception of excluded paths).

  • When using this mode, there can be no conflicts—the contents of target always override the contents of source.

two-way-safe (or alias two-way)

  • Bidirectionally syncs a local source to a remote target path.

  • Changes made in the local source will be synced to the remote target.

  • Changes made in the remote target will be synced to the local source.

  • When there are conflicts on either side, does not replace/delete the corresponding conflicting paths on the other side.

  • Similarly to one-way, this mode is simple to configure when there are files in either source or target that you don't want overridden on the other side when files change or are added/deleted.

  • Setting up several one-way-replica and one-way-replica-reverse syncs instead of one-way and two-way is generally the best approach long-term, but may require more fine-grained configuration (more sync specs for specific subpaths and more specific exclusion rules, to make sure things don't get overwritten/deleted in unwanted ways).

two-way-resolved

Same as two-way-safe except:

  • Changes made in the local source will always win any conflict. This includes cases where alpha’s deletions would overwrite beta’s modifications or creations

  • No conflicts can occur in this synchronization mode.

In addition to the above, please check out the Mutagen docs on synchronization for more info.

Notes on Mutagen terminology

Mutagen uses the terminology "alpha" and "beta" for the sync endpoints. In Garden's one-way, one-way-replica and two-way sync modes, alpha is source and beta is target.

For the reverse sync modes (one-way-reverse and one-way-replica-reverse), alpha is target and beta is source.

Excluding files and directories from syncs

By design, Garden's dev mode does not apply exclusion rules from ignorefiles (such as .gardenignore files) to dev mode syncs.

This is done to grant you more control over precisely which files and directories you'd like to sync while in dev mode.

For example, you might want to ignore dist or build directories (not version control them, not include them in builds or module versions), but still be able to sync them from your local machine to the running container (or from the running container to your local machine). This is easy to achieve with dev mode.

Exclusion rules can be specified on individual sync configs:

kind: Module
name: node-service
type: container
services:
  - name: node-service
    args: [npm, start]
    devMode:
      command: [npm, run, dev]
      sync:
        - source: src
          target: /app/src
          exclude: [node_modules, tmp, "**/*.log"] # <------ paths matching these patterns won't be synced
          mode: two-way
  ...

Project-wide exclusion rules can be set on the local-kubernetes and kubernetes providers:

kind: Project
...
providers:
  - name: kubernetes
    ...
    # Configure project-wide exclusion rules and default permission/ownership settings
    # for synced files/directories.
    devMode:
      defaults:
        exclude:
          - "/**/node_modules" # <--- with this, we don't have to specify `node_modules` on individual sync specs

This is great to reduce repetition in your excludes.

See the reference documentation for the kubernetes provider) for a full list of provider-level options for dev mode when using the kubernetes provider. The same dev-mode options are also available when using local-kubernetes.

Permissions and ownership

In certain cases you may need to set a specific owner/group or permission bits on the synced files and directories at the target.

To do this, you can set a few options on each sync:

kind: Module
description: Node greeting service
name: node-service
type: container
services:
  - name: node-service
    args: [npm, start]
    devMode:
      command: [npm, run, dev]
      sync:
        - target: /app
          exclude: [node_modules]
          defaultOwner: 1000  # <- set an integer user ID or a string name
          defaultGroup: 1000  # <- set an integer group ID or a string name
          defaultFileMode: 0666  # <- set the permission bits (as octals) for synced files
          defaultDirectoryMode: 0777  # <- set the permission bits (as octals) for synced directories
  ...

These options are passed directly to Mutagen. For more information, please see the Mutagen docs.

An advanced example

This example demonstrates several of the more advanced options that dev mode offers. For more details on the options available, see the sections above.

kind: Project
...
providers:
  - name: kubernetes
    ...
    # Configure project-wide exclusion rules and default permission/ownership settings
    # for synced files/directories.
    devMode:
      defaults:
        exclude:
          - "/**/node_modules"
        owner: 1000  # <- set an integer user ID or a string name
        group: 1000  # <- set an integer group ID or a string name
        fileMode: 0666  # <- set the permission bits (as octals) for synced files
        directoryMode: 0777  # <- set the permission bits (as octals) for synced directories

---

kind: Module
description: |
  Here, we sync source code into the remote, and sync back the `test-artifacts` directory
  (populated when we run tests) back to the local machine.
name: node-service
type: container
services:
  - name: node-service
    args: [npm, start]
    devMode:
      command: [npm, run, dev] # Overrides the container's default when the service is deployed in dev mode.
      sync:
        # You can use several sync specs for the same service. It's generally a good idea to be specific about
        # what you want to sync, and to use `one-way-replica` or `one-way-replica-reverse` when possible to keep
        # things simple and avoid sync conflicts.
        - source: app
          target: /app
          # We don't need to exclude `node_modules` here, since above we added a
          # project-wide exclusion rule for that.
          # exclude: [node_modules]
          mode: one-way-replica
        - source: test-artifacts
          target: /test-artifacts
          # This syncs back any files/folders  on the remote to the local machine, always
          # overriding the local directory's contents with the remote one. See above for a detailed
          # description of each available sync mode.
          mode: one-way-replica-reverse
  ...

Troubleshooting

Every so often something comes up in the underlying Mutagen synchronization process, which may not be visible in the Garden CLI logs. To figure out what the issue may be (say, ahead of reporting a GitHub issue for Garden), it's useful to be able to use the mutagen CLI directly.

Because Garden creates a temporary data directory for Mutagen for every Garden CLI instance, you can't use the mutagen CLI without additional context. However, to make this easier, a symlink to the temporary directory is automatically created under <project root>/.garden/mutagen/<random ID>, as well as a mutagen.sh helper script within that directory that sets the appropriate context and links to the automatically installed Mutagen CLI. We also create a <project root>/.garden/mutagen/latest symlink for convenience.

To, for example, get the current list of active syncs in an active Garden process, you could run the following from the project root directory:

.garden/mutagen/latest/mutagen.sh sync list

Last updated