Skip to content

rs_dpr_service/dask/dask_cluster_handler.md

<< Back to index

Module to handle connection to Dask cluster.

DaskClusterHandler

Class to handle connection to Dask cluster.

NOTE: a new instance of this class is called for every endpoint call.

Source code in docs/rs-dpr-service/rs_dpr_service/dask/dask_cluster_handler.py 
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
class DaskClusterHandler:  # pylint: disable=too-few-public-methods
    """
    Class to handle connection to Dask cluster.

    NOTE: a new instance of this class is called for every endpoint call.
    """

    def __init__(self, cluster_info: ClusterInfo, local_mode_address: str):
        self.cluster_info = cluster_info
        self.cluster_address = os.environ[local_mode_address] if LOCAL_MODE else os.environ["DASK_GATEWAY_ADDRESS"]
        self.cluster: GatewayCluster

    def _connect_to_cluster(self):
        """
        Handles the first part of setup_dask_connection.
        See there for details.
        """
        cluster_label = self.cluster_info.cluster_label

        # Connect to the gateway and get the list of the clusters
        try:
            # In local mode, authenticate to the dask cluster with username/password
            if LOCAL_MODE:
                gateway_auth = BasicAuth(
                    os.environ["LOCAL_DASK_USERNAME"],
                    os.environ["LOCAL_DASK_PASSWORD"],
                )

            # Cluster mode
            else:
                # check the auth type, only jupyterhub type supported for now
                auth_type = os.environ["DASK_GATEWAY__AUTH__TYPE"]
                # Handle JupyterHub authentication
                if auth_type == "jupyterhub":
                    gateway_auth = JupyterHubAuth(api_token=self.cluster_info.jupyter_token)
                else:
                    logger.error(f"Unsupported authentication type: {auth_type}")
                    raise RuntimeError(f"Unsupported authentication type: {auth_type}")

            gateway = Gateway(
                address=self.cluster_address,
                auth=gateway_auth,
            )

            # Sort the clusters by newest first
            clusters = sorted(gateway.list_clusters(), key=lambda cluster: cluster.start_time, reverse=True)
            logger.debug(f"Cluster list for gateway {self.cluster_address!r}: {clusters}")

            # We need to find the cluster instance, if it is not set in the input info
            if not self.cluster_info.cluster_instance:

                # In local mode, get the first cluster from the gateway.
                # This cluster instance id is needed by the eopf dask scheduler to connect later to this cluster.
                # This is something like "dask-gateway.17e196069443463495547eb97f532834"
                if LOCAL_MODE:
                    if clusters:
                        self.cluster_info.cluster_instance = clusters[0].name

                # In cluster mode, get the instance of the cluster identified by its label.
                else:
                    logger.info(f"Cluster label: {cluster_label}")

                    for cluster in clusters:
                        logger.info(f"Existing cluster labels: {cluster.options.get('cluster_name')}")

                        is_equal = cluster.options.get("cluster_name") == cluster_label
                        logger.info(f"Is equal: {is_equal}")

                    self.cluster_info.cluster_instance = next(
                        (
                            cluster.name
                            for cluster in clusters
                            if isinstance(cluster.options, dict)
                            and cluster.options.get("cluster_name") == cluster_label
                        ),
                        "",
                    )
                    logger.info(f"Cluster instance: {self.cluster_info.cluster_instance}")

                if not self.cluster_info.cluster_instance:
                    raise IndexError(f"Dask cluster with 'cluster_name'={cluster_label!r} was not found.")

            self.cluster = gateway.connect(self.cluster_info.cluster_instance)
            if not self.cluster:
                logger.exception("Failed to create the cluster")
                raise RuntimeError("Failed to create the cluster")
            logger.info(f"Successfully connected to the {cluster_label!r} dask cluster")

        except KeyError as e:
            logger.exception(
                "Failed to retrieve the required connection details for "
                "the Dask Gateway from one or more of the following environment variables: "
                "DASK_GATEWAY_ADDRESS, RSPY_DASK_DPR_SERVICE_CLUSTER_NAME, "
                f"DASK_GATEWAY__AUTH__TYPE. {e}",
            )

            raise RuntimeError(
                f"Failed to retrieve the required connection details for Dask Gateway. Missing key:{e}",
            ) from e
        except IndexError as e:
            logger.exception(f"Failed to find the specified dask cluster: {e}")
            raise RuntimeError(f"No dask cluster named {cluster_label!r} was found.") from e

    def setup_dask_connection(self) -> Client:
        """Connects a dask cluster scheduler
        Establishes a connection to a Dask cluster, either in a local environment or via a Dask Gateway in
        a Kubernetes cluster. This method checks if the cluster is already created (for local mode) or connects
        to a Dask Gateway to find or create a cluster scheduler (for Kubernetes mode, see RSPY_LOCAL_MODE env var).

        1. **Local Mode**:
        - If `self.cluster` already exists, it assumes the Dask cluster was created when the application started,
            and proceeds without creating a new cluster.

        2. **Kubernetes Mode**:
        - If `self.cluster` is not already defined, the method attempts to connect to a Dask Gateway
            (using environment variables `DASK_GATEWAY_ADDRESS` and `DASK_GATEWAY__AUTH__TYPE`) to
            retrieve a list of existing clusters.
        - If no clusters are available, it attempts to create a new cluster scheduler.

        Raises:
            RuntimeError: Raised if the cluster name is None, required environment variables are missing,
                        cluster creation fails or authentication errors occur.
            KeyError: Raised if the necessary Dask Gateway environment variables (`DASK_GATEWAY_ADDRESS`,
                `DASK_GATEWAY__AUTH__TYPE`, `RSPY_DASK_DPR_SERVICE_CLUSTER_NAME` ) are not set.
            IndexError: Raised if no clusters are found in the Dask Gateway and new cluster creation is attempted.
            dask_gateway.exceptions.GatewayServerError: Raised when there is a server-side error in Dask Gateway.
            dask_gateway.exceptions.AuthenticationError: Raised if authentication to the Dask Gateway fails.
            dask_gateway.exceptions.ClusterLimitExceeded: Raised if the limit on the number of clusters is exceeded.

        Behavior:
        1. **Cluster Creation and Connection**:
            - In Kubernetes mode, the method tries to connect to an existing cluster or creates
            a new one if none exists.
            - Error handling includes catching issues like missing environment variables, authentication failures,
            cluster creation timeouts, or exceeding cluster limits.

        2. **Logging**:
            - Logs the list of available clusters if connected via the Dask Gateway.
            - Logs the success of the connection or any errors encountered during the process.
            - Logs the Dask dashboard URL and the number of active workers.

        3. **Client Initialization**:
            - Once connected to the Dask cluster, the method creates a Dask `Client` object for managing tasks
            and logs the number of running workers.
            - If no workers are found, it scales the cluster to 1 worker.

        4. **Error Handling**:
            - Handles various exceptions during the connection and creation process, including:
            - Missing environment variables.
            - Failures during cluster creation.
            - Issues related to cluster scaling, worker retrieval, or client creation.
            - If an error occurs, the method logs the error and attempts to gracefully handle failure.

        Returns:
            Dask client
        """
        self._connect_to_cluster()

        logger.debug("Cluster dashboard: %s", self.cluster.dashboard_link)
        # create the client as well
        client = Client(self.cluster)

        # Forward logging from dask workers to the caller
        client.forward_logging()

        # Upload local module to the dask client.
        call_dask.upload_this_module(client)

        # set_dask_env function is in utils, uploaded to the dask cluster in call_dask
        client.run(set_dask_env, os.environ)

        # This is a temporary fix for the dask cluster settings which does not create a scheduler by default
        # This code should be removed as soon as this is fixed in the kubernetes cluster
        try:
            logger.debug(f"{client.get_versions(check=True)}")
            workers = client.scheduler_info()["workers"]
            logger.info(f"Number of running workers: {len(workers)}")

        except Exception as e:  # pylint: disable=broad-exception-caught
            logger.exception(f"Dask cluster client failed: {e}")
            raise RuntimeError(f"Dask cluster client failed: {e}") from e
        if len(workers) == 0:
            logger.info("No workers are currently running in the Dask cluster. Scaling up to 1.")
            self.cluster.scale(1)

        # Check the cluster dashboard
        logger.debug(f"Dask Client: {client} | Cluster dashboard: {self.cluster.dashboard_link}")

        return client

setup_dask_connection()

Connects a dask cluster scheduler Establishes a connection to a Dask cluster, either in a local environment or via a Dask Gateway in a Kubernetes cluster. This method checks if the cluster is already created (for local mode) or connects to a Dask Gateway to find or create a cluster scheduler (for Kubernetes mode, see RSPY_LOCAL_MODE env var).

  1. Local Mode: - If self.cluster already exists, it assumes the Dask cluster was created when the application started, and proceeds without creating a new cluster.

  2. Kubernetes Mode: - If self.cluster is not already defined, the method attempts to connect to a Dask Gateway (using environment variables DASK_GATEWAY_ADDRESS and DASK_GATEWAY__AUTH__TYPE) to retrieve a list of existing clusters. - If no clusters are available, it attempts to create a new cluster scheduler.

Raises:

Type Description
RuntimeError

Raised if the cluster name is None, required environment variables are missing, cluster creation fails or authentication errors occur.
KeyError

Raised if the necessary Dask Gateway environment variables (DASK_GATEWAY_ADDRESS, DASK_GATEWAY__AUTH__TYPE, RSPY_DASK_DPR_SERVICE_CLUSTER_NAME ) are not set.
IndexError

Raised if no clusters are found in the Dask Gateway and new cluster creation is attempted.
GatewayServerError

Raised when there is a server-side error in Dask Gateway.
AuthenticationError

Raised if authentication to the Dask Gateway fails.
ClusterLimitExceeded

Raised if the limit on the number of clusters is exceeded.

Behavior: 1. Cluster Creation and Connection: - In Kubernetes mode, the method tries to connect to an existing cluster or creates a new one if none exists. - Error handling includes catching issues like missing environment variables, authentication failures, cluster creation timeouts, or exceeding cluster limits.

  1. Logging:

    • Logs the list of available clusters if connected via the Dask Gateway.
    • Logs the success of the connection or any errors encountered during the process.
    • Logs the Dask dashboard URL and the number of active workers.

  2. Client Initialization:

    • Once connected to the Dask cluster, the method creates a Dask Client object for managing tasks and logs the number of running workers.
    • If no workers are found, it scales the cluster to 1 worker.

  3. Error Handling:

    • Handles various exceptions during the connection and creation process, including:
    • Missing environment variables.
    • Failures during cluster creation.
    • Issues related to cluster scaling, worker retrieval, or client creation.
    • If an error occurs, the method logs the error and attempts to gracefully handle failure.

Returns:

Type Description
Client

Dask client
Source code in docs/rs-dpr-service/rs_dpr_service/dask/dask_cluster_handler.py 
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
def setup_dask_connection(self) -> Client:
    """Connects a dask cluster scheduler
    Establishes a connection to a Dask cluster, either in a local environment or via a Dask Gateway in
    a Kubernetes cluster. This method checks if the cluster is already created (for local mode) or connects
    to a Dask Gateway to find or create a cluster scheduler (for Kubernetes mode, see RSPY_LOCAL_MODE env var).

    1. **Local Mode**:
    - If `self.cluster` already exists, it assumes the Dask cluster was created when the application started,
        and proceeds without creating a new cluster.

    2. **Kubernetes Mode**:
    - If `self.cluster` is not already defined, the method attempts to connect to a Dask Gateway
        (using environment variables `DASK_GATEWAY_ADDRESS` and `DASK_GATEWAY__AUTH__TYPE`) to
        retrieve a list of existing clusters.
    - If no clusters are available, it attempts to create a new cluster scheduler.

    Raises:
        RuntimeError: Raised if the cluster name is None, required environment variables are missing,
                    cluster creation fails or authentication errors occur.
        KeyError: Raised if the necessary Dask Gateway environment variables (`DASK_GATEWAY_ADDRESS`,
            `DASK_GATEWAY__AUTH__TYPE`, `RSPY_DASK_DPR_SERVICE_CLUSTER_NAME` ) are not set.
        IndexError: Raised if no clusters are found in the Dask Gateway and new cluster creation is attempted.
        dask_gateway.exceptions.GatewayServerError: Raised when there is a server-side error in Dask Gateway.
        dask_gateway.exceptions.AuthenticationError: Raised if authentication to the Dask Gateway fails.
        dask_gateway.exceptions.ClusterLimitExceeded: Raised if the limit on the number of clusters is exceeded.

    Behavior:
    1. **Cluster Creation and Connection**:
        - In Kubernetes mode, the method tries to connect to an existing cluster or creates
        a new one if none exists.
        - Error handling includes catching issues like missing environment variables, authentication failures,
        cluster creation timeouts, or exceeding cluster limits.

    2. **Logging**:
        - Logs the list of available clusters if connected via the Dask Gateway.
        - Logs the success of the connection or any errors encountered during the process.
        - Logs the Dask dashboard URL and the number of active workers.

    3. **Client Initialization**:
        - Once connected to the Dask cluster, the method creates a Dask `Client` object for managing tasks
        and logs the number of running workers.
        - If no workers are found, it scales the cluster to 1 worker.

    4. **Error Handling**:
        - Handles various exceptions during the connection and creation process, including:
        - Missing environment variables.
        - Failures during cluster creation.
        - Issues related to cluster scaling, worker retrieval, or client creation.
        - If an error occurs, the method logs the error and attempts to gracefully handle failure.

    Returns:
        Dask client
    """
    self._connect_to_cluster()

    logger.debug("Cluster dashboard: %s", self.cluster.dashboard_link)
    # create the client as well
    client = Client(self.cluster)

    # Forward logging from dask workers to the caller
    client.forward_logging()

    # Upload local module to the dask client.
    call_dask.upload_this_module(client)

    # set_dask_env function is in utils, uploaded to the dask cluster in call_dask
    client.run(set_dask_env, os.environ)

    # This is a temporary fix for the dask cluster settings which does not create a scheduler by default
    # This code should be removed as soon as this is fixed in the kubernetes cluster
    try:
        logger.debug(f"{client.get_versions(check=True)}")
        workers = client.scheduler_info()["workers"]
        logger.info(f"Number of running workers: {len(workers)}")

    except Exception as e:  # pylint: disable=broad-exception-caught
        logger.exception(f"Dask cluster client failed: {e}")
        raise RuntimeError(f"Dask cluster client failed: {e}") from e
    if len(workers) == 0:
        logger.info("No workers are currently running in the Dask cluster. Scaling up to 1.")
        self.cluster.scale(1)

    # Check the cluster dashboard
    logger.debug(f"Dask Client: {client} | Cluster dashboard: {self.cluster.dashboard_link}")

    return client