HPCC Systems Git support

Gavin Halliday (Enterprise/Lead Architect, LexisNexis Risk Solutions) has a wealth of knowledge when it comes to everything relating to the ECL language. He is one of the original architects of the HPCC System platform and has worked for almost 3 decades with the LexisNexis Risk Solutions Group. In this blog, Gavin focuses on some new features available in GitHub and shares his knowledge about the best way to implement them. These features will greatly increase the speed and usability of GitHub.

For many years, HPCC Systems has used the eclccserver to compile ECL code from GitHub repositories. The repositories (and optional branches/users) are configured using environment variables on the server. By configuring the Githook mechanism, you can submit a query from a branch and the eclccserver will pull the source from a GitHub or GitLab repository and compile it. This allows you to deploy a specific version of a query to production without needing to perform any work on the client side.

Recent versions of the platform have significantly improved support for compiling from the Git repositories. Only some of the changes have been included in our older support releases, so we recommend that you run on the most updated version to take advantage of the full potential of all the features now available.

Git improvements now available in HPCC Systems

These are the significant changes that are now available in versions of the platform since HPCC Systems 7.12.x:

Speed improvements

The code for interfacing with Git repositories has been significantly improved. On large queries, this can reduce the compile time by 10s of minutes. (Previously it ran the Git command as a child process. Now it uses a library to read the repositories directly.)

Resources and manifests from Git

This is a significant bug fix. Previously, manifests and resources were only associated with a query if they were contained in source files, which meant they were not included when compiling using the Githook mechanism.

Support for Git-lfs

Git-lfs is an extension to Git that improves support for large files and is supported by GitHub and GitLab. If the Git lfs extension is installed then eclccserver will retrieve files from Git-lfs. This is particularly useful for large resources, for example, java packages included as part of the manifest.

Note: You must update to a recent point release to gain these improvements. The speed improvements alone are a compelling reason to ensure you are using the most up to date version.

The problem with large projects

The Githook mechanism works well when the source code is all contained within a single repository. This is not an issue for small to medium projects, but a single repository doesn’t scale so well for large projects. For example, you might have one team providing common utility code shared between many teams, another responsible for data ingest and index generation and a third team creating queries using that data. There are also likely to be other teams, for example, those developing external tools like SALT/KEL, and even bundles.

So how do you manage these independent source streams? Previously there were a couple of options.

Use a single repository

All teams are working on a single repository. Each of the teams is likely to be working on different products with different schedules. Each team needs to maintain a separate set of branches for the different versions of their code (within the same repository), and changes need to be periodically merged between branches owned by the different teams.

Managing multiple semi-independent branches in the same repo is a recipe for confusion. Also, some code from external tools is guaranteed to be managed in a different repo.

Merging repositories

Each team works on their own repository and merges changes in from other repositories as they are needed. The advantage of this is that the branch structure within a repository is simpler. Integrating changes from other teams is probably worse though.

How do you ensure all the changes from the other team are taken? How do you avoid problems with incompatible changes or dependencies? This approach is likely to lead to multiple copies of the same source not really being kept in sync.This is the problem that the new Git support aims to solve.

New Git features available in HPCC Systems 8.6.x

While some of the following features are also available HPCC Systems 8.4.x, the most recent changes can be found in HPCC Systems 8.6.x., so it is our recommendation that you use the latest available point release.

Compiling from a Git repository

To compile a query directly from a repository the –main syntax has been extended. You can now use the command:

ecl run thor --main demo.main@https://github.com/ghalliday/gch-ecldemo-d#version1 --server=...

This submits a query to Thor via ESP. It will retrieve ECL code from the ‘version1’ branch in the repository, compile the code in file demo/main.ecl file and run the query on Thor. Similar to the old Githhook mechanism, the checkout will be done on the remote eclccserver rather than on the client machine.

The reference to the repository location has the following form:


If the protocol and urn are omitted, a default will be used, so this example command can be simplified as follows:

ecl run thor --main demo.main@ghalliday/gch-ecldemo-d#version1 --server=...

The version-text that follows the # can take one of the following forms:

  • The name of a branch
  • The name of a tag
  • A SHA of a commit

The branch ‘version1’ in ghalliday/gch-ecldemo-d has the SHA as 3c23ca0, so the following command would be equivalent:

ecl run thor --main demo.main@ghalliday/gch-ecldemo-d#3c23ca0 --server=...

If the version1 branch changed, the first command would compile the updated branch while the second form would always compile the same code.

This feature is also supported for local compiles using eclcc. So the following command locally syntax checks the same code:

eclcc --main demo.main@ghalliday/gch-ecldemo-d#3c23ca0 -syntax

But, this time the source is downloaded and processed on the client machine.

Packages and dependencies

The other changes in HPCC Systems 8.6.x aim to resolve the following issues:

  1. To allow queries to be compiled from multiple repositories, avoiding problems with clashing symbols.
    This is a problem when merging source code from multiple teams. Clashes can occur if the teams have modules with the same name, and other errors if you are using the legacy import semantics.
  2. To ensure that the dependencies between repositories are versioned.

The HPCC Systems platform improvements allow each Git repository to be treated as a separate independent package. Dependencies between the repositories are specified in a package file which is checked into the repository and versioned along with the ECL code. The package file indicates what the dependencies are and which versions should be used. Many of the ideas are the same as those used by the node package manager npm.

This is best illustrated using the following example for package.json:

  "name": "demoRepoC",
  "version": "1.0.0", 
  "dependencies": {
    "demoRepoD": "ghalliday/gch-ecldemo-d#version1"

At its simplest, the package file gives a name to the package and defines the dependencies. The dependencies property is a list of key-value pairs. The key provides the name of the ECL module that is used to access the external repository. The value is a repository reference which uses the same format as a repository in the –main syntax.

To use definitions from the external repository you need to add an import definition to your ECL code. For example, you might have the following format/personAsText.ecl file:

IMPORT layout;
IMPORT demoRepoD AS demoD;

EXPORT personAsText(layout.person input) :=
    input.name + ': ' + demoD.format.maskPassword(input.password);

The name demoRepoD in the second import matches the key value in the package.json file. This code uses the attribute format.maskPassword from the repository ghalliday/gch-ecldemo-d .

Each package is processed independently of any others. The only connection is through explicit imports of the external packages. This means that packages can have modules or attributes with the same name and they will not clash with each other.

It is even possible (although not encouraged) to use multiple versions of an external package.

An example of that for package.json:

  "name": "demoRepoC",
  "version": "1.0.0",
  "dependencies": {
    "demoRepoD_V1": "ghalliday/gch-ecldemo-d#version1"
    "demoRepoD_V2": "ghalliday/gch-ecldemo-d#version2"

And an example for query.ecl:

IMPORT layout;
IMPORT demoRepoD_V1 AS demo1;
IMPORT demoRepoD_V2 AS demo2;

EXPORT personAsText(layout.person input) :=
    'Was: ' + demo1.format.maskPassword(input.password) +
    ' Now: ' + demo2.format.maskPassword(input.password);

This is useful if different teams have dependencies on different versions of a third party repository. It is possible to use code from both teams without having to resolve their dependencies first. (You are likely to get smaller queries if only single version of a repository is being used.)

Local development aids

There are a couple of options added to the eclcc and ecl commands to aid ECL developers. One common situation is where the developer is working on multiple repositories at the same time. The -R option indicates that instead of using source from an external repository, the compiler should use it from a local directory.

The syntax is as follows:


For example:

ecl run examples/main.ecl -Rghalliday/gch-ecldemo-d=/home/myuser/source/demod

This will then use ECL code for DemoRepoD from /home/myuser/source/demoD rather than https://github.com/ghalliday/gch-ecldemo-d#version1.

If you have any issues not resolving repositories, it can be helpful to use the -v option to enable verbose logging output, including details of the Git requests.

Advanced options

The following options are available.

Configuring the use of Git within Helm charts for the cloud version

  • eclccserver.gitUsername
  • secrets.git
    See the Authentication section below for more details.
  • eclccserver.gitPlane
    The eclccserver instances are ephemeral, but the same repositories tend to be shared by queries, so it makes sense to be able to cache and share the cloned packages between instances. This options allows the user to define the storage plane that external packages are cloned to.
    If the option is not supplied, the default is the first storage plane with a category of Git – otherwise eclccserver uses the first storage plane with a category of dll.

Options for ECL and eclcc:

  • –defaultgitprefix
    This command line option changes the default prefix that is added to relative packages references. The default can also be configured using the environment variable ECLCC_DEFAULT_GITPREFIX. It defaults to “https://github.com/“.
  • –fetchrepos
    Should external repos that have not been cloned locally be fetched? This defaults to true in 8.6.x. It may be useful to set this option to false if all external repos are mapped to local directories to double check they are being redirected correctly.
  • –updaterepos
    Should external repos that have previously been fetched locally be updated? This option defaults to true. It is useful to set this option to false if you are working in a situation with no access to the external repositories, or to avoid the overhead of checking for changes if you know there are none.
    Which directory are the external repos cloned to? On a client machine this defaults to <home>/.HPCCSystems/repos (or <APPDATA>\HPCCSystems\repos on windows). You can delete the contents of this directory to force a clean download of all repositories.

npm and package-lock.json

For most developers using the package.json files to define the dependencies will be sufficient. However, it is possible to use the npm package manager to ensure that labels or branches are tied down to a specific SHA.

For example:

npm install --package-lock-only

This command creates a package-lock.json file in the same location as package.json. The npm program will resolve the references to branches and resolve them to the corresponding SHAs. (The –package-lock-only option indicates that npm should not clone the associated versions of the code to node_modules directories.) The generated package-lock.json file will contain something similar to the following:

  "packages": {
    "node_modules/demoRepoD": {
      "resolved": "https://github.com/ghalliday/gch-ecldemo-d.git#644c4585221f4dd80ca1e8f05974983455a244e5",

If a package-lock.json file is present it will take precedence over the package.json file. The ECL is compiled in the same way, with eclcc automatically downloading the dependencies.

What is the advantage of using package-lock.json over package.json? It allows you to use npm’s semantic versioning syntax (#semver) which is not currently supported by eclcc. It also allows you to use a branch in your package.json file as a logical dependency, but resolved to an actual dependency on specific SHA, so that if the branch is updated the query will not change.

Finally npm install can also be used without the –package-lock-only option. This will check out the appropriate version of the code into the node_modules subdirectory of the current project. eclcc supports the node_module structure as a way of providing the source for external packages. This is an alternative way to compile the code using eclcc completely independently from the source control system.

Migrating existing code

If you have an existing single repository and you want to split it up into multiple repositories, what is the best way to go about it?

This is best illustrated using an example. Let’s look at extracting common utilities (currently in a ut module) in to a separate repository.

The original example.ecl code that uses it might look like this:

EXPORT example(string request, string user) := ut.normalizeRequest(user + ':' + request);

The aim is to extract all of the ut module into another repository, and modify the example code to use it.

Creating the package.json

First, define a package file indicating your example code is now dependent on a new external repository:

  "name": "myExampleRepo",
  "version": "1.0.0",
  "dependencies": {
    "utilities": "myorg/shared-utilities#version1"

Move all the utility code into the ‘myorg/shared-utilities’ repository, and check it in as the version1 branch. Delete the utility code from the original location.


Next, change the remaining code to use that imported external package:

IMPORT utilities.ut;
EXPORT example(string request, string user) := ut.normalizeRequest(user + ':' + request);


IMPORT utilities;
EXPORT example(string request, string user) := utilities.ut.normalizeRequest(user + ':' + request);

If you are starting from scratch, this would probably be the clearest way to express the imports. However, it can be painful to have go through all the existing source replacing ‘IMPORT ut;’ with ‘IMPORT utilities.ut;’. One alternative is to define an attribute in the example repository which defines an alias for a module from an external repository. In this case, that means creating a ut.ecl file within the root of the main repository as follows:

IMPORT utilities;
EXPORT ut := utilities.ut;

Using this definition, all exiting imports of ut will continue to work and will now refer to the code in the external repository, while your existing source code can remain the same. It is worth noting that one consequence of using this approach is that it prevents you having modules with the same name in different repositories.


Macros differ from other ECL attribute definitions because they are expanded as source in the context they are called from. This means any identifiers are interpreted in the calling context, instead of the context the macro is defined in. This can cause problems for imports within macros.

For example, if you have an attribute for processing all fields in a ut.processing.processFile.ecl file like this:

EXPORT processFile(infile) := FUNCTION MACRO
   IMPORT ut;

If this macro is then moved into a separate repository, when it is next called, the ‘IMPORT ut;’ will be processed in the context of the caller’s repository. That will work if the repository contains an alias for ut, but otherwise the import will fail. The solution is to use the #\$ symbol in the macro. This symbol expands to the module or folder that contains the definition of the macro (in this case ut.processing), including the context of the repository in which it was defined. The following definition will work from wherever it is called:

EXPORT processFile(infile) := FUNCTION MACRO
   IMPORT #$.^ AS ut;

Where #\$ corresponds to ut.processing, so #\$.^ corresponds to ut.


Some of these features overlap with the idea of ECL bundles, which are packages of ECL code that can be installed and used from your own ECL code. However, there some significant differences:

  • Imports within bundles need to be relative
  • Bundles need to be installed before they can be used
  • There is no way to create a dependency on a particular version of a bundle. It depends on what is installed

The new Git package mechanism can be used to included existing bundles. So instead of installing a bundle using the following command;

ecl bundle install DataPatterns

Use the following package.json file to do it for you:

  "dependencies": {
    "DataPatterns": "hpcc-systems/DataPatterns#v1.8.2"

The advantage of this approach is that there is a clear dependency on the version and the bundle does not need to be separately installed on the eclccserver machine.

Note: One minor implication of the new features is that the archive format has been extended. Each external package will be included as a nested archive with a package tag to indicate which package it corresponds to. (All external archives are contained within the root archive, and external packages will only be included once.)


If the external repositories are public (e.g. bundles) then there are no futher requirements. Private repositories have the additional complication of requiring authentication information – either on the client on eclccserver depending on where the source is gathered. Git provides various methods for providing the credentials…

Credentials for client machines or local development

The following are the recommended approaches for configuring the credentials on a local development system that is interacting with GitHub:

  • ssh key In this scenario, the ssh key associated with the local developer machine is registered with the GitHub account. For more details click here. This is used when the GitHub reference is of the form ssh://github.com. The sshkey can be protected with a passcode, and there are various options to avoid having to enter the passcode each time.
  • github authentication Download the GitHub command line tool. You can then use it to authenticate all Git access with the following command:
gh auth login
  • This is probably the simplest option if you are using github. More details can be found here
  • Use a personal access token
    These are similar to a password, but with additional restrictions on their lifetime and the resources that can be accessed. Here are the details on how to to create them. They can then be used with the various git credential caching options. Here is an example.

Note: It is preferrable to use the https:// protocol instead of ssh:// for links in package-lock.json files. If ssh:// is used it requires any machine that processes the dependency to have access to a registered ssh key.

All of the options above are likely to involve some user interaction, such as passphrases for ssh keys, web interaction with github authentication, and initial entry for cached access tokens. This is problematic for the eclccserver which cannot support user interaction, and it is preferrable not to pass credentials around.

The solution is to use a personal access token securely stored as a secret. (This token would generally be associated with a special service account.) The secret avoids the need to pass credentials and allows the keys to be rotated.

The following sections describe secrets support in the Kubernetes and Bare Metal versions of the HPCC Systems platform.

Credentials for eclccserver – Kubernetes Version

In Kubernetes you need to take the following steps:

  1. Add the gitUsername property to the eclccserver component in the value.yaml file:
- name: myeclccserver
  gitUsername: ghalliday
  1. Add a secret to the values.yaml file, with a key that matches the username:
    ghalliday: my-git-secret
  1. Add a secret to Kubernetes containing the personal access token:
apiVersion: v1
kind: Secret
  name: my-git-secret
type: Opaque
  password: ghp_eZLHeuoHxxxxxxxxxxxxxxxxxxxxol3986sS=
kubectl apply -f ~/dev/hpcc/helm/secrets/my-git-secret

When a query is submitted to eclccserver, any git repositories are accessed using the user name and password configured above.

  1. Store the secret in a vault:

You can also store the PAT inside a vault.  See the documentation and values.yaml file for details of how to configure vaults in more detail.

Credentials for eclccserver – Bare-metal

Bare-metal systems require some similar configuration steps:

First you will define the environment variable HPCC_GIT_USERNAME

Add the gitUsername property to the EclCCServerProcess entry in the environment file.

export HPCC_GIT_USERNAME=ghalliday

<EclCCServerProcess daliServers="mydali"

Then you will need to do one of the two options below:

  1. Store the access token in /opt/HPCCSystems/secrets/git/<user-name>/password

For example:

$cat /opt/HPCCSystems/secrets/git/ghalliday/password


  1. You can store the credentials in a vault.

Within the Environment/Software section you can now define a vaults section.

For example:

    <git name='my-storage-vault' url="${secret}" 
kind="kv-v2" client-secret="myVaultSecret"/>

The entries have the same content as the entries in the kubernetes value.yaml file. The xml above corresponds to the yaml content:

     - name: 'my-storage-vault'
       url: "${secret}"
       kind: "kv-v2"
       client-secret: "myVaultSecret"

Future development

A number of additional features and options to aid developers are being considered for future versions. In a future blog, we will be going back over file storage in kubernetes in order too revisit and consolidate the changes that have happened over the last 2 years.

Semantic versioning

npm supports a special tag of #semver to allow semantic versioning to be used. Currently, you can only use this if you use the package-lock.json files generated by npm install. It is possible that this will be supported natively in the future.

New Support for dependencies between multiple repositories

This can be used to track the progress of any other proposed features. See this JIRA issue for more information.