diff --git a/contributing/BACKENDS.md b/contributing/BACKENDS.md
index 7a7a742316..f53750791a 100644
--- a/contributing/BACKENDS.md
+++ b/contributing/BACKENDS.md
@@ -42,7 +42,7 @@ base class. See its docstrings for descriptions of the methods that your class s
Refer to examples:
- Offline providers:
- [datacrunch.py](https://github.com/dstackai/gpuhunt/blob/main/src/gpuhunt/providers/datacrunch.py),
+ [verda.py](https://github.com/dstackai/gpuhunt/blob/main/src/gpuhunt/providers/verda.py),
[aws.py](https://github.com/dstackai/gpuhunt/blob/main/src/gpuhunt/providers/aws.py),
[azure.py](https://github.com/dstackai/gpuhunt/blob/main/src/gpuhunt/providers/azure.py),
[lambdalabs.py](https://github.com/dstackai/gpuhunt/blob/main/src/gpuhunt/providers/lambdalabs.py).
@@ -64,7 +64,7 @@ Add your provider in the following places:
For offline providers, you can add data quality tests under `src/integrity_tests/`.
Data quality tests are run after collecting offline catalogs to ensure their integrity.
-Refer to examples: [test_datacrunch.py](https://github.com/dstackai/gpuhunt/blob/main/src/integrity_tests/test_datacrunch.py),
+Refer to examples: [test_verda.py](https://github.com/dstackai/gpuhunt/blob/main/src/integrity_tests/test_verda.py),
[test_gcp.py](https://github.com/dstackai/gpuhunt/blob/main/src/integrity_tests/test_gcp.py).
### 1.6. Submit a pull request
@@ -125,7 +125,7 @@ Then add these models to `AnyBackendConfig*` unions in [`src/dstack/_internal/co
The script also generates `*BackendStoredConfig` that extends `*BackendConfig` to be able to store extra parameters in the DB. By the same logic, it generates `*Config` that extends `*BackendStoredConfig` with creds and uses it as the main `Backend` and `Compute` config instead of using `*BackendConfigWithCreds` directly.
Refer to examples:
-[datacrunch](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/datacrunch/models.py),
+[verda](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/verda/models.py),
[aws](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/aws/models.py),
[gcp](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/gcp/models.py),
[azure](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/models.py), etc.
@@ -136,7 +136,7 @@ Go to `compute.py` and implement `Compute` methods.
Optionally, extend and implement `ComputeWith*` classes to support additional features such as fleets, volumes, gateways, placement groups, etc. For example, extend `ComputeWithCreateInstanceSupport` to support fleets.
Refer to examples:
-[datacrunch](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/datacrunch/compute.py),
+[verda](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/verda/compute.py),
[aws](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/aws/compute.py),
[gcp](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/gcp/compute.py),
[azure](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/azure/compute.py), etc.
@@ -146,7 +146,7 @@ Refer to examples:
Go to `configurator.py` and implement custom `Configurator` logic. At minimum, you should implement creds validation.
You may also need to validate other config parameters if there are any.
-Refer to examples: [datacrunch](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/datacrunch/configurator.py),
+Refer to examples: [verda](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/verda/configurator.py),
[aws](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/aws/configurator.py),
[gcp](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/gcp/configurator.py),
[azure](https://github.com/dstackai/dstack/blob/master/src/dstack/_internal/core/backends/azure/configurator.py), etc.
@@ -195,7 +195,7 @@ For some VM-based backends, the `dstack` team also maintains
[custom VM images](../scripts/packer/README.md) with the required dependencies
and `dstack`-specific optimizations.
-Examples of VM-based backends include: `aws`, `azure`, `gcp`, `lambda`, `datacrunch`, `tensordock`, etc.
+Examples of VM-based backends include: `aws`, `azure`, `gcp`, `lambda`, `verda`, etc.
#### 3.1.2. Container-based backend compute type
diff --git a/contributing/GPUHUNT.md b/contributing/GPUHUNT.md
index c30ccc807a..33e3150fa1 100644
--- a/contributing/GPUHUNT.md
+++ b/contributing/GPUHUNT.md
@@ -65,7 +65,7 @@ Some providers offer extreme flexibility in possible configurations, but not all
- Filters out if: outdated family, not supported family
- Queries configuration details to fill CPU, RAM, and GPU information
-### DataCrunch
+### Verda
- Just queries all offers via API
diff --git a/docs/assets/images/verda-logo.svg b/docs/assets/images/verda-logo.svg
new file mode 100644
index 0000000000..3f688a7d44
--- /dev/null
+++ b/docs/assets/images/verda-logo.svg
@@ -0,0 +1,13 @@
+
+
+
+
+
+
+
diff --git a/docs/docs/concepts/backends.md b/docs/docs/concepts/backends.md
index 7c1eb0b791..574342318f 100644
--- a/docs/docs/concepts/backends.md
+++ b/docs/docs/concepts/backends.md
@@ -854,9 +854,11 @@ There are two ways to configure OCI: using client credentials or using the defau
compartment_id: ocid1.compartment.oc1..aaaaaaaa
```
-### DataCrunch
+
-
- DataCrunch
+ Verda
diff --git a/pyproject.toml b/pyproject.toml
index 5ad026de64..0d149339d7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -32,7 +32,7 @@ dependencies = [
"python-multipart>=0.0.16",
"filelock",
"psutil",
- "gpuhunt==0.1.14",
+ "gpuhunt==0.1.15",
"argcomplete>=3.5.0",
"ignore-python>=0.2.0",
"orjson",
@@ -186,7 +186,11 @@ gcp = [
"dstack[server]",
]
datacrunch = [
- "datacrunch",
+ "verda; python_version >= '3.10'",
+ "dstack[server]",
+]
+verda = [
+ "verda; python_version >= '3.10'",
"dstack[server]",
]
kubernetes = [
@@ -211,5 +215,5 @@ nebius = [
"dstack[server]",
]
all = [
- "dstack[gateway,server,aws,azure,gcp,datacrunch,kubernetes,lambda,nebius,oci]",
+ "dstack[gateway,server,aws,azure,gcp,verda,kubernetes,lambda,nebius,oci]",
]
diff --git a/src/dstack/_internal/core/backends/base/offers.py b/src/dstack/_internal/core/backends/base/offers.py
index de1d7c4875..d4fae91af5 100644
--- a/src/dstack/_internal/core/backends/base/offers.py
+++ b/src/dstack/_internal/core/backends/base/offers.py
@@ -39,6 +39,8 @@ def get_catalog_offers(
catalog: Optional[gpuhunt.Catalog] = None,
) -> List[InstanceOffer]:
provider = backend.value
+ if backend == BackendType.DATACRUNCH:
+ provider = BackendType.VERDA.value # Backward compatibility
if backend == BackendType.LAMBDA:
provider = "lambdalabs"
if backend == BackendType.AMDDEVCLOUD:
diff --git a/src/dstack/_internal/core/backends/configurators.py b/src/dstack/_internal/core/backends/configurators.py
index eeb91d0bac..ec7f976c53 100644
--- a/src/dstack/_internal/core/backends/configurators.py
+++ b/src/dstack/_internal/core/backends/configurators.py
@@ -127,6 +127,15 @@
except ImportError:
pass
+try:
+ from dstack._internal.core.backends.verda.configurator import (
+ VerdaConfigurator,
+ )
+
+ _CONFIGURATOR_CLASSES.append(VerdaConfigurator)
+except ImportError:
+ pass
+
try:
from dstack._internal.core.backends.vultr.configurator import VultrConfigurator
diff --git a/src/dstack/_internal/core/backends/datacrunch/__init__.py b/src/dstack/_internal/core/backends/datacrunch/__init__.py
index e69de29bb2..ca4773c861 100644
--- a/src/dstack/_internal/core/backends/datacrunch/__init__.py
+++ b/src/dstack/_internal/core/backends/datacrunch/__init__.py
@@ -0,0 +1 @@
+# DataCrunch backend for backward compatibility
diff --git a/src/dstack/_internal/core/backends/datacrunch/backend.py b/src/dstack/_internal/core/backends/datacrunch/backend.py
index 7df591c289..1ce1c97c42 100644
--- a/src/dstack/_internal/core/backends/datacrunch/backend.py
+++ b/src/dstack/_internal/core/backends/datacrunch/backend.py
@@ -1,16 +1,18 @@
-from dstack._internal.core.backends.base.backend import Backend
from dstack._internal.core.backends.datacrunch.compute import DataCrunchCompute
-from dstack._internal.core.backends.datacrunch.models import DataCrunchConfig
+from dstack._internal.core.backends.verda.backend import VerdaBackend
+from dstack._internal.core.backends.verda.models import VerdaConfig
from dstack._internal.core.models.backends.base import BackendType
-class DataCrunchBackend(Backend):
+# Deprecated
+# TODO: Remove in 0.21
+class DataCrunchBackend(VerdaBackend):
TYPE = BackendType.DATACRUNCH
COMPUTE_CLASS = DataCrunchCompute
- def __init__(self, config: DataCrunchConfig):
+ def __init__(self, config: VerdaConfig):
self.config = config
- self._compute = DataCrunchCompute(self.config)
+ self._compute = DataCrunchCompute(self.config, self.TYPE)
def compute(self) -> DataCrunchCompute:
return self._compute
diff --git a/src/dstack/_internal/core/backends/datacrunch/compute.py b/src/dstack/_internal/core/backends/datacrunch/compute.py
index 6543567c39..906c9ea2e5 100644
--- a/src/dstack/_internal/core/backends/datacrunch/compute.py
+++ b/src/dstack/_internal/core/backends/datacrunch/compute.py
@@ -1,265 +1,8 @@
-from collections.abc import Iterable
-from typing import Dict, List, Optional
-
-from datacrunch import DataCrunchClient
-from datacrunch.exceptions import APIException
-from datacrunch.instances.instances import Instance
-
-from dstack._internal.core.backends.base.backend import Compute
-from dstack._internal.core.backends.base.compute import (
- ComputeWithAllOffersCached,
- ComputeWithCreateInstanceSupport,
- ComputeWithPrivilegedSupport,
- generate_unique_instance_name,
- get_shim_commands,
-)
-from dstack._internal.core.backends.base.offers import (
- OfferModifier,
- get_catalog_offers,
- get_offers_disk_modifier,
-)
-from dstack._internal.core.backends.datacrunch.models import DataCrunchConfig
-from dstack._internal.core.errors import NoCapacityError
+from dstack._internal.core.backends.verda.compute import VerdaCompute
+from dstack._internal.core.backends.verda.models import VerdaConfig
from dstack._internal.core.models.backends.base import BackendType
-from dstack._internal.core.models.instances import (
- InstanceAvailability,
- InstanceConfiguration,
- InstanceOffer,
- InstanceOfferWithAvailability,
-)
-from dstack._internal.core.models.placement import PlacementGroup
-from dstack._internal.core.models.resources import Memory, Range
-from dstack._internal.core.models.runs import JobProvisioningData, Requirements
-from dstack._internal.utils.logging import get_logger
-from dstack._internal.utils.ssh import get_public_key_fingerprint
-
-logger = get_logger("datacrunch.compute")
-
-MAX_INSTANCE_NAME_LEN = 60
-
-IMAGE_SIZE = Memory.parse("50GB")
-
-CONFIGURABLE_DISK_SIZE = Range[Memory](min=IMAGE_SIZE, max=None)
-
-
-class DataCrunchCompute(
- ComputeWithAllOffersCached,
- ComputeWithCreateInstanceSupport,
- ComputeWithPrivilegedSupport,
- Compute,
-):
- def __init__(self, config: DataCrunchConfig):
- super().__init__()
- self.config = config
- self.client = DataCrunchClient(
- client_id=self.config.creds.client_id,
- client_secret=self.config.creds.client_secret,
- )
-
- def get_all_offers_with_availability(self) -> List[InstanceOfferWithAvailability]:
- offers = get_catalog_offers(
- backend=BackendType.DATACRUNCH,
- locations=self.config.regions,
- )
- offers_with_availability = self._get_offers_with_availability(offers)
- return offers_with_availability
-
- def get_offers_modifiers(self, requirements: Requirements) -> Iterable[OfferModifier]:
- return [get_offers_disk_modifier(CONFIGURABLE_DISK_SIZE, requirements)]
-
- def _get_offers_with_availability(
- self, offers: List[InstanceOffer]
- ) -> List[InstanceOfferWithAvailability]:
- raw_availabilities: List[Dict] = self.client.instances.get_availabilities()
-
- region_availabilities = {}
- for location in raw_availabilities:
- location_code = location["location_code"]
- availabilities = location["availabilities"]
- for name in availabilities:
- key = (name, location_code)
- region_availabilities[key] = InstanceAvailability.AVAILABLE
-
- availability_offers = []
- for offer in offers:
- key = (offer.instance.name, offer.region)
- availability = region_availabilities.get(key, InstanceAvailability.NOT_AVAILABLE)
- availability_offers.append(
- InstanceOfferWithAvailability(**offer.dict(), availability=availability)
- )
-
- return availability_offers
-
- def create_instance(
- self,
- instance_offer: InstanceOfferWithAvailability,
- instance_config: InstanceConfiguration,
- placement_group: Optional[PlacementGroup],
- ) -> JobProvisioningData:
- instance_name = generate_unique_instance_name(
- instance_config, max_length=MAX_INSTANCE_NAME_LEN
- )
- public_keys = instance_config.get_public_keys()
- ssh_ids = []
- for ssh_public_key in public_keys:
- ssh_ids.append(
- # datacrunch allows you to use the same name
- _get_or_create_ssh_key(
- client=self.client,
- name=f"dstack-{instance_config.instance_name}.key",
- public_key=ssh_public_key,
- )
- )
-
- commands = get_shim_commands(authorized_keys=public_keys)
- startup_script = " ".join([" && ".join(commands)])
- script_name = f"dstack-{instance_config.instance_name}.sh"
- startup_script_ids = _get_or_create_startup_scrpit(
- client=self.client,
- name=script_name,
- script=startup_script,
- )
-
- disk_size = round(instance_offer.instance.resources.disk.size_mib / 1024)
- image_id = _get_vm_image_id(instance_offer)
-
- logger.debug(
- "Deploying datacrunch instance",
- {
- "instance_type": instance_offer.instance.name,
- "ssh_key_ids": ssh_ids,
- "startup_script_id": startup_script_ids,
- "hostname": instance_name,
- "description": instance_name,
- "image": image_id,
- "disk_size": disk_size,
- "location": instance_offer.region,
- },
- )
- instance = _deploy_instance(
- client=self.client,
- instance_type=instance_offer.instance.name,
- ssh_key_ids=ssh_ids,
- startup_script_id=startup_script_ids,
- hostname=instance_name,
- description=instance_name,
- image=image_id,
- disk_size=disk_size,
- is_spot=instance_offer.instance.resources.spot,
- location=instance_offer.region,
- )
- return JobProvisioningData(
- backend=instance_offer.backend,
- instance_type=instance_offer.instance,
- instance_id=instance.id,
- hostname=None,
- internal_ip=None,
- region=instance.location,
- price=instance_offer.price,
- username="root",
- ssh_port=22,
- dockerized=True,
- ssh_proxy=None,
- backend_data=None,
- )
-
- def terminate_instance(
- self, instance_id: str, region: str, backend_data: Optional[str] = None
- ):
- try:
- self.client.instances.action(id_list=[instance_id], action="delete")
- except APIException as e:
- if e.message in [
- "Invalid instance id",
- "Can't discontinue a discontinued instance",
- ]:
- logger.debug("Skipping instance %s termination. Instance not found.", instance_id)
- return
- raise
-
- def update_provisioning_data(
- self,
- provisioning_data: JobProvisioningData,
- project_ssh_public_key: str,
- project_ssh_private_key: str,
- ):
- instance = _get_instance_by_id(self.client, provisioning_data.instance_id)
- if instance is not None and instance.status == "running":
- provisioning_data.hostname = instance.ip
-
-
-def _get_vm_image_id(instance_offer: InstanceOfferWithAvailability) -> str:
- # https://api.datacrunch.io/v1/images
- if len(instance_offer.instance.resources.gpus) > 0 and instance_offer.instance.resources.gpus[
- 0
- ].name in ["V100", "A6000"]:
- # Ubuntu 22.04 + CUDA 12.0 + Docker
- return "2088da25-bb0d-41cc-a191-dccae45d96fd"
- # Ubuntu 24.04 + CUDA 12.8 Open + Docker
- return "77777777-4f48-4249-82b3-f199fb9b701b"
-
-
-def _get_or_create_ssh_key(client: DataCrunchClient, name: str, public_key: str) -> str:
- fingerprint = get_public_key_fingerprint(public_key)
- keys = client.ssh_keys.get()
- found_keys = [key for key in keys if fingerprint == get_public_key_fingerprint(key.public_key)]
- if found_keys:
- key = found_keys[0]
- return key.id
- key = client.ssh_keys.create(name, public_key)
- return key.id
-
-
-def _get_or_create_startup_scrpit(client: DataCrunchClient, name: str, script: str) -> str:
- scripts = client.startup_scripts.get()
- found_scripts = [startup_script for startup_script in scripts if script == startup_script]
- if found_scripts:
- startup_script = found_scripts[0]
- return startup_script.id
-
- startup_script = client.startup_scripts.create(name, script)
- return startup_script.id
-
-
-def _get_instance_by_id(
- client: DataCrunchClient,
- instance_id: str,
-) -> Optional[Instance]:
- try:
- return client.instances.get_by_id(instance_id)
- except APIException as e:
- if e.message == "Invalid instance id":
- return None
- raise
-
-def _deploy_instance(
- client: DataCrunchClient,
- instance_type: str,
- image: str,
- ssh_key_ids: List[str],
- hostname: str,
- description: str,
- startup_script_id: str,
- disk_size: int,
- is_spot: bool,
- location: str,
-) -> Instance:
- try:
- instance = client.instances.create(
- instance_type=instance_type,
- image=image,
- ssh_key_ids=ssh_key_ids,
- hostname=hostname,
- description=description,
- startup_script_id=startup_script_id,
- pricing="FIXED_PRICE",
- is_spot=is_spot,
- location=location,
- os_volume={"name": "OS volume", "size": disk_size},
- )
- except APIException as e:
- # FIXME: Catch only no capacity errors
- raise NoCapacityError(f"DataCrunch API error: {e.message}")
- return instance
+class DataCrunchCompute(VerdaCompute):
+ def __init__(self, config: VerdaConfig, backend_type: BackendType):
+ super().__init__(config, backend_type)
diff --git a/src/dstack/_internal/core/backends/datacrunch/configurator.py b/src/dstack/_internal/core/backends/datacrunch/configurator.py
index f31d5a69a1..944f8657d3 100644
--- a/src/dstack/_internal/core/backends/datacrunch/configurator.py
+++ b/src/dstack/_internal/core/backends/datacrunch/configurator.py
@@ -1,77 +1,17 @@
-import json
-
-from datacrunch import DataCrunchClient
-from datacrunch.exceptions import APIException
-
-from dstack._internal.core.backends.base.configurator import (
- BackendRecord,
- Configurator,
- raise_invalid_credentials_error,
-)
+from dstack._internal.core.backends.base.configurator import BackendRecord
from dstack._internal.core.backends.datacrunch.backend import DataCrunchBackend
-from dstack._internal.core.backends.datacrunch.models import (
- DataCrunchBackendConfig,
- DataCrunchBackendConfigWithCreds,
- DataCrunchConfig,
- DataCrunchCreds,
- DataCrunchStoredConfig,
+from dstack._internal.core.backends.verda.configurator import (
+ VerdaConfigurator,
)
from dstack._internal.core.models.backends.base import (
BackendType,
)
-class DataCrunchConfigurator(
- Configurator[
- DataCrunchBackendConfig,
- DataCrunchBackendConfigWithCreds,
- ]
-):
+class DataCrunchConfigurator(VerdaConfigurator):
TYPE = BackendType.DATACRUNCH
BACKEND_CLASS = DataCrunchBackend
- def validate_config(
- self, config: DataCrunchBackendConfigWithCreds, default_creds_enabled: bool
- ):
- self._validate_creds(config.creds)
-
- def create_backend(
- self, project_name: str, config: DataCrunchBackendConfigWithCreds
- ) -> BackendRecord:
- return BackendRecord(
- config=DataCrunchStoredConfig(
- **DataCrunchBackendConfig.__response__.parse_obj(config).dict()
- ).json(),
- auth=DataCrunchCreds.parse_obj(config.creds).json(),
- )
-
- def get_backend_config_with_creds(
- self, record: BackendRecord
- ) -> DataCrunchBackendConfigWithCreds:
- config = self._get_config(record)
- return DataCrunchBackendConfigWithCreds.__response__.parse_obj(config)
-
- def get_backend_config_without_creds(self, record: BackendRecord) -> DataCrunchBackendConfig:
- config = self._get_config(record)
- return DataCrunchBackendConfig.__response__.parse_obj(config)
-
def get_backend(self, record: BackendRecord) -> DataCrunchBackend:
config = self._get_config(record)
return DataCrunchBackend(config=config)
-
- def _get_config(self, record: BackendRecord) -> DataCrunchConfig:
- return DataCrunchConfig.__response__(
- **json.loads(record.config),
- creds=DataCrunchCreds.parse_raw(record.auth),
- )
-
- def _validate_creds(self, creds: DataCrunchCreds):
- try:
- DataCrunchClient(
- client_id=creds.client_id,
- client_secret=creds.client_secret,
- )
- except APIException as e:
- if e.code == "unauthorized_request":
- raise_invalid_credentials_error(fields=[["creds", "api_key"]])
- raise
diff --git a/src/dstack/_internal/core/backends/datacrunch/models.py b/src/dstack/_internal/core/backends/datacrunch/models.py
deleted file mode 100644
index 5a98e4cc82..0000000000
--- a/src/dstack/_internal/core/backends/datacrunch/models.py
+++ /dev/null
@@ -1,38 +0,0 @@
-from typing import Annotated, List, Literal, Optional, Union
-
-from pydantic import Field
-
-from dstack._internal.core.models.common import CoreModel
-
-
-class DataCrunchAPIKeyCreds(CoreModel):
- type: Annotated[Literal["api_key"], Field(description="The type of credentials")] = "api_key"
- client_id: Annotated[str, Field(description="The client ID")]
- client_secret: Annotated[str, Field(description="The client secret")]
-
-
-AnyDataCrunchCreds = DataCrunchAPIKeyCreds
-DataCrunchCreds = AnyDataCrunchCreds
-
-
-class DataCrunchBackendConfig(CoreModel):
- type: Annotated[Literal["datacrunch"], Field(description="The type of backend")] = "datacrunch"
- regions: Annotated[
- Optional[List[str]],
- Field(description="The list of DataCrunch regions. Omit to use all regions"),
- ] = None
-
-
-class DataCrunchBackendConfigWithCreds(DataCrunchBackendConfig):
- creds: Annotated[AnyDataCrunchCreds, Field(description="The credentials")]
-
-
-AnyDataCrunchBackendConfig = Union[DataCrunchBackendConfig, DataCrunchBackendConfigWithCreds]
-
-
-class DataCrunchStoredConfig(DataCrunchBackendConfig):
- pass
-
-
-class DataCrunchConfig(DataCrunchStoredConfig):
- creds: AnyDataCrunchCreds
diff --git a/src/dstack/_internal/core/backends/models.py b/src/dstack/_internal/core/backends/models.py
index 1715080f83..f1c59e2f44 100644
--- a/src/dstack/_internal/core/backends/models.py
+++ b/src/dstack/_internal/core/backends/models.py
@@ -16,10 +16,6 @@
CudoBackendConfig,
CudoBackendConfigWithCreds,
)
-from dstack._internal.core.backends.datacrunch.models import (
- DataCrunchBackendConfig,
- DataCrunchBackendConfigWithCreds,
-)
from dstack._internal.core.backends.digitalocean_base.models import (
BaseDigitalOceanBackendConfig,
BaseDigitalOceanBackendConfigWithCreds,
@@ -68,6 +64,10 @@
VastAIBackendConfig,
VastAIBackendConfigWithCreds,
)
+from dstack._internal.core.backends.verda.models import (
+ VerdaBackendConfig,
+ VerdaBackendConfigWithCreds,
+)
from dstack._internal.core.backends.vultr.models import (
VultrBackendConfig,
VultrBackendConfigWithCreds,
@@ -80,7 +80,6 @@
AzureBackendConfig,
CloudRiftBackendConfig,
CudoBackendConfig,
- DataCrunchBackendConfig,
BaseDigitalOceanBackendConfig,
GCPBackendConfig,
HotAisleBackendConfig,
@@ -91,6 +90,7 @@
RunpodBackendConfig,
TensorDockBackendConfig,
VastAIBackendConfig,
+ VerdaBackendConfig,
VultrBackendConfig,
DstackBackendConfig,
DstackBaseBackendConfig,
@@ -104,7 +104,7 @@
AzureBackendConfigWithCreds,
CloudRiftBackendConfigWithCreds,
CudoBackendConfigWithCreds,
- DataCrunchBackendConfigWithCreds,
+ VerdaBackendConfigWithCreds,
BaseDigitalOceanBackendConfigWithCreds,
GCPBackendConfigWithCreds,
HotAisleBackendConfigWithCreds,
@@ -127,7 +127,7 @@
AzureBackendConfigWithCreds,
CloudRiftBackendConfigWithCreds,
CudoBackendConfigWithCreds,
- DataCrunchBackendConfigWithCreds,
+ VerdaBackendConfigWithCreds,
BaseDigitalOceanBackendConfigWithCreds,
GCPBackendFileConfigWithCreds,
HotAisleBackendFileConfigWithCreds,
diff --git a/src/tests/_internal/core/backends/datacrunch/__init__.py b/src/dstack/_internal/core/backends/verda/__init__.py
similarity index 100%
rename from src/tests/_internal/core/backends/datacrunch/__init__.py
rename to src/dstack/_internal/core/backends/verda/__init__.py
diff --git a/src/dstack/_internal/core/backends/verda/backend.py b/src/dstack/_internal/core/backends/verda/backend.py
new file mode 100644
index 0000000000..eed2beb34d
--- /dev/null
+++ b/src/dstack/_internal/core/backends/verda/backend.py
@@ -0,0 +1,16 @@
+from dstack._internal.core.backends.base.backend import Backend
+from dstack._internal.core.backends.verda.compute import VerdaCompute
+from dstack._internal.core.backends.verda.models import VerdaConfig
+from dstack._internal.core.models.backends.base import BackendType
+
+
+class VerdaBackend(Backend):
+ TYPE = BackendType.VERDA
+ COMPUTE_CLASS = VerdaCompute
+
+ def __init__(self, config: VerdaConfig):
+ self.config = config
+ self._compute = VerdaCompute(self.config, self.TYPE)
+
+ def compute(self) -> VerdaCompute:
+ return self._compute
diff --git a/src/dstack/_internal/core/backends/verda/compute.py b/src/dstack/_internal/core/backends/verda/compute.py
new file mode 100644
index 0000000000..d6dbdd6ae0
--- /dev/null
+++ b/src/dstack/_internal/core/backends/verda/compute.py
@@ -0,0 +1,266 @@
+from collections.abc import Iterable
+from typing import Dict, List, Optional
+
+from verda import VerdaClient
+from verda.exceptions import APIException
+from verda.instances import Instance
+
+from dstack._internal.core.backends.base.backend import Compute
+from dstack._internal.core.backends.base.compute import (
+ ComputeWithAllOffersCached,
+ ComputeWithCreateInstanceSupport,
+ ComputeWithPrivilegedSupport,
+ generate_unique_instance_name,
+ get_shim_commands,
+)
+from dstack._internal.core.backends.base.offers import (
+ OfferModifier,
+ get_catalog_offers,
+ get_offers_disk_modifier,
+)
+from dstack._internal.core.backends.verda.models import VerdaConfig
+from dstack._internal.core.errors import NoCapacityError
+from dstack._internal.core.models.backends.base import BackendType
+from dstack._internal.core.models.instances import (
+ InstanceAvailability,
+ InstanceConfiguration,
+ InstanceOffer,
+ InstanceOfferWithAvailability,
+)
+from dstack._internal.core.models.placement import PlacementGroup
+from dstack._internal.core.models.resources import Memory, Range
+from dstack._internal.core.models.runs import JobProvisioningData, Requirements
+from dstack._internal.utils.logging import get_logger
+from dstack._internal.utils.ssh import get_public_key_fingerprint
+
+logger = get_logger("verda.compute")
+
+MAX_INSTANCE_NAME_LEN = 60
+
+IMAGE_SIZE = Memory.parse("50GB")
+
+CONFIGURABLE_DISK_SIZE = Range[Memory](min=IMAGE_SIZE, max=None)
+
+
+class VerdaCompute(
+ ComputeWithAllOffersCached,
+ ComputeWithCreateInstanceSupport,
+ ComputeWithPrivilegedSupport,
+ Compute,
+):
+ def __init__(self, config: VerdaConfig, backend_type: BackendType):
+ super().__init__()
+ self.config = config
+ self.client = VerdaClient(
+ client_id=self.config.creds.client_id,
+ client_secret=self.config.creds.client_secret,
+ )
+ self.backend_type = backend_type
+
+ def get_all_offers_with_availability(self) -> List[InstanceOfferWithAvailability]:
+ offers = get_catalog_offers(
+ backend=self.backend_type,
+ locations=self.config.regions,
+ )
+ offers_with_availability = self._get_offers_with_availability(offers)
+ return offers_with_availability
+
+ def get_offers_modifiers(self, requirements: Requirements) -> Iterable[OfferModifier]:
+ return [get_offers_disk_modifier(CONFIGURABLE_DISK_SIZE, requirements)]
+
+ def _get_offers_with_availability(
+ self, offers: List[InstanceOffer]
+ ) -> List[InstanceOfferWithAvailability]:
+ raw_availabilities: List[Dict] = self.client.instances.get_availabilities()
+
+ region_availabilities = {}
+ for location in raw_availabilities:
+ location_code = location["location_code"]
+ availabilities = location["availabilities"]
+ for name in availabilities:
+ key = (name, location_code)
+ region_availabilities[key] = InstanceAvailability.AVAILABLE
+
+ availability_offers = []
+ for offer in offers:
+ key = (offer.instance.name, offer.region)
+ availability = region_availabilities.get(key, InstanceAvailability.NOT_AVAILABLE)
+ availability_offers.append(
+ InstanceOfferWithAvailability(**offer.dict(), availability=availability)
+ )
+
+ return availability_offers
+
+ def create_instance(
+ self,
+ instance_offer: InstanceOfferWithAvailability,
+ instance_config: InstanceConfiguration,
+ placement_group: Optional[PlacementGroup],
+ ) -> JobProvisioningData:
+ instance_name = generate_unique_instance_name(
+ instance_config, max_length=MAX_INSTANCE_NAME_LEN
+ )
+ public_keys = instance_config.get_public_keys()
+ ssh_ids = []
+ for ssh_public_key in public_keys:
+ ssh_ids.append(
+ # verda allows you to use the same name
+ _get_or_create_ssh_key(
+ client=self.client,
+ name=f"dstack-{instance_config.instance_name}.key",
+ public_key=ssh_public_key,
+ )
+ )
+
+ commands = get_shim_commands(authorized_keys=public_keys)
+ startup_script = " ".join([" && ".join(commands)])
+ script_name = f"dstack-{instance_config.instance_name}.sh"
+ startup_script_ids = _get_or_create_startup_scrpit(
+ client=self.client,
+ name=script_name,
+ script=startup_script,
+ )
+
+ disk_size = round(instance_offer.instance.resources.disk.size_mib / 1024)
+ image_id = _get_vm_image_id(instance_offer)
+
+ logger.debug(
+ "Deploying Verda instance",
+ {
+ "instance_type": instance_offer.instance.name,
+ "ssh_key_ids": ssh_ids,
+ "startup_script_id": startup_script_ids,
+ "hostname": instance_name,
+ "description": instance_name,
+ "image": image_id,
+ "disk_size": disk_size,
+ "location": instance_offer.region,
+ },
+ )
+ instance = _deploy_instance(
+ client=self.client,
+ instance_type=instance_offer.instance.name,
+ ssh_key_ids=ssh_ids,
+ startup_script_id=startup_script_ids,
+ hostname=instance_name,
+ description=instance_name,
+ image=image_id,
+ disk_size=disk_size,
+ is_spot=instance_offer.instance.resources.spot,
+ location=instance_offer.region,
+ )
+ return JobProvisioningData(
+ backend=instance_offer.backend,
+ instance_type=instance_offer.instance,
+ instance_id=instance.id,
+ hostname=None,
+ internal_ip=None,
+ region=instance.location,
+ price=instance_offer.price,
+ username="root",
+ ssh_port=22,
+ dockerized=True,
+ ssh_proxy=None,
+ backend_data=None,
+ )
+
+ def terminate_instance(
+ self, instance_id: str, region: str, backend_data: Optional[str] = None
+ ):
+ try:
+ self.client.instances.action(id_list=[instance_id], action="delete")
+ except APIException as e:
+ if e.message in [
+ "Invalid instance id",
+ "Can't discontinue a discontinued instance",
+ ]:
+ logger.debug("Skipping instance %s termination. Instance not found.", instance_id)
+ return
+ raise
+
+ def update_provisioning_data(
+ self,
+ provisioning_data: JobProvisioningData,
+ project_ssh_public_key: str,
+ project_ssh_private_key: str,
+ ):
+ instance = _get_instance_by_id(self.client, provisioning_data.instance_id)
+ if instance is not None and instance.status == "running":
+ provisioning_data.hostname = instance.ip
+
+
+def _get_vm_image_id(instance_offer: InstanceOfferWithAvailability) -> str:
+ # https://api.verda.com/v1/images
+ if len(instance_offer.instance.resources.gpus) > 0 and instance_offer.instance.resources.gpus[
+ 0
+ ].name in ["V100", "A6000"]:
+ # Ubuntu 22.04 + CUDA 12.0 + Docker
+ return "2088da25-bb0d-41cc-a191-dccae45d96fd"
+ # Ubuntu 24.04 + CUDA 12.8 Open + Docker
+ return "77777777-4f48-4249-82b3-f199fb9b701b"
+
+
+def _get_or_create_ssh_key(client: VerdaClient, name: str, public_key: str) -> str:
+ fingerprint = get_public_key_fingerprint(public_key)
+ keys = client.ssh_keys.get()
+ found_keys = [key for key in keys if fingerprint == get_public_key_fingerprint(key.public_key)]
+ if found_keys:
+ key = found_keys[0]
+ return key.id
+ key = client.ssh_keys.create(name, public_key)
+ return key.id
+
+
+def _get_or_create_startup_scrpit(client: VerdaClient, name: str, script: str) -> str:
+ scripts = client.startup_scripts.get()
+ found_scripts = [startup_script for startup_script in scripts if script == startup_script]
+ if found_scripts:
+ startup_script = found_scripts[0]
+ return startup_script.id
+
+ startup_script = client.startup_scripts.create(name, script)
+ return startup_script.id
+
+
+def _get_instance_by_id(
+ client: VerdaClient,
+ instance_id: str,
+) -> Optional[Instance]:
+ try:
+ return client.instances.get_by_id(instance_id)
+ except APIException as e:
+ if e.message == "Invalid instance id":
+ return None
+ raise
+
+
+def _deploy_instance(
+ client: VerdaClient,
+ instance_type: str,
+ image: str,
+ ssh_key_ids: List[str],
+ hostname: str,
+ description: str,
+ startup_script_id: str,
+ disk_size: int,
+ is_spot: bool,
+ location: str,
+) -> Instance:
+ try:
+ instance = client.instances.create(
+ instance_type=instance_type,
+ image=image,
+ ssh_key_ids=ssh_key_ids,
+ hostname=hostname,
+ description=description,
+ startup_script_id=startup_script_id,
+ pricing="FIXED_PRICE",
+ is_spot=is_spot,
+ location=location,
+ os_volume={"name": "OS volume", "size": disk_size},
+ )
+ except APIException as e:
+ # FIXME: Catch only no capacity errors
+ raise NoCapacityError(f"Verda API error: {e.message}")
+
+ return instance
diff --git a/src/dstack/_internal/core/backends/verda/configurator.py b/src/dstack/_internal/core/backends/verda/configurator.py
new file mode 100644
index 0000000000..64b0dec034
--- /dev/null
+++ b/src/dstack/_internal/core/backends/verda/configurator.py
@@ -0,0 +1,73 @@
+import json
+
+from verda import VerdaClient
+from verda.exceptions import APIException
+
+from dstack._internal.core.backends.base.configurator import (
+ BackendRecord,
+ Configurator,
+ raise_invalid_credentials_error,
+)
+from dstack._internal.core.backends.verda.backend import VerdaBackend
+from dstack._internal.core.backends.verda.models import (
+ VerdaBackendConfig,
+ VerdaBackendConfigWithCreds,
+ VerdaConfig,
+ VerdaCreds,
+ VerdaStoredConfig,
+)
+from dstack._internal.core.models.backends.base import (
+ BackendType,
+)
+
+
+class VerdaConfigurator(
+ Configurator[
+ VerdaBackendConfig,
+ VerdaBackendConfigWithCreds,
+ ]
+):
+ TYPE = BackendType.VERDA
+ BACKEND_CLASS = VerdaBackend
+
+ def validate_config(self, config: VerdaBackendConfigWithCreds, default_creds_enabled: bool):
+ self._validate_creds(config.creds)
+
+ def create_backend(
+ self, project_name: str, config: VerdaBackendConfigWithCreds
+ ) -> BackendRecord:
+ return BackendRecord(
+ config=VerdaStoredConfig(
+ **VerdaBackendConfig.__response__.parse_obj(config).dict()
+ ).json(),
+ auth=VerdaCreds.parse_obj(config.creds).json(),
+ )
+
+ def get_backend_config_with_creds(self, record: BackendRecord) -> VerdaBackendConfigWithCreds:
+ config = self._get_config(record)
+ return VerdaBackendConfigWithCreds.__response__.parse_obj(config)
+
+ def get_backend_config_without_creds(self, record: BackendRecord) -> VerdaBackendConfig:
+ config = self._get_config(record)
+ return VerdaBackendConfig.__response__.parse_obj(config)
+
+ def get_backend(self, record: BackendRecord) -> VerdaBackend:
+ config = self._get_config(record)
+ return VerdaBackend(config=config)
+
+ def _get_config(self, record: BackendRecord) -> VerdaConfig:
+ return VerdaConfig.__response__(
+ **json.loads(record.config),
+ creds=VerdaCreds.parse_raw(record.auth),
+ )
+
+ def _validate_creds(self, creds: VerdaCreds):
+ try:
+ VerdaClient(
+ client_id=creds.client_id,
+ client_secret=creds.client_secret,
+ )
+ except APIException as e:
+ if e.code == "unauthorized_request":
+ raise_invalid_credentials_error(fields=[["creds", "api_key"]])
+ raise
diff --git a/src/dstack/_internal/core/backends/verda/models.py b/src/dstack/_internal/core/backends/verda/models.py
new file mode 100644
index 0000000000..1e0b896b5d
--- /dev/null
+++ b/src/dstack/_internal/core/backends/verda/models.py
@@ -0,0 +1,38 @@
+from typing import Annotated, List, Literal, Optional, Union
+
+from pydantic import Field
+
+from dstack._internal.core.models.common import CoreModel
+
+
+class VerdaAPIKeyCreds(CoreModel):
+ type: Annotated[Literal["api_key"], Field(description="The type of credentials")] = "api_key"
+ client_id: Annotated[str, Field(description="The client ID")]
+ client_secret: Annotated[str, Field(description="The client secret")]
+
+
+AnyVerdaCreds = VerdaAPIKeyCreds
+VerdaCreds = AnyVerdaCreds
+
+
+class VerdaBackendConfig(CoreModel):
+ type: Annotated[Literal["verda", "datacrunch"], Field(description="The type of backend")]
+ regions: Annotated[
+ Optional[List[str]],
+ Field(description="The list of Verda regions. Omit to use all regions"),
+ ] = None
+
+
+class VerdaBackendConfigWithCreds(VerdaBackendConfig):
+ creds: Annotated[AnyVerdaCreds, Field(description="The credentials")]
+
+
+AnyVerdaBackendConfig = Union[VerdaBackendConfig, VerdaBackendConfigWithCreds]
+
+
+class VerdaStoredConfig(VerdaBackendConfig):
+ pass
+
+
+class VerdaConfig(VerdaStoredConfig):
+ creds: AnyVerdaCreds
diff --git a/src/dstack/_internal/core/models/backends/base.py b/src/dstack/_internal/core/models/backends/base.py
index 3d33e75b62..82efe09efa 100644
--- a/src/dstack/_internal/core/models/backends/base.py
+++ b/src/dstack/_internal/core/models/backends/base.py
@@ -9,7 +9,7 @@ class BackendType(str, enum.Enum):
AZURE (BackendType): Microsoft Azure
CLOUDRIFT (BackendType): CloudRift
CUDO (BackendType): Cudo
- DATACRUNCH (BackendType): DataCrunch
+ DATACRUNCH (BackendType): DataCrunch (for backward compatibility)
DIGITALOCEAN (BackendType): DigitalOcean
DSTACK (BackendType): dstack Sky
GCP (BackendType): Google Cloud Platform
@@ -21,6 +21,7 @@ class BackendType(str, enum.Enum):
RUNPOD (BackendType): Runpod Cloud
TENSORDOCK (BackendType): TensorDock Marketplace
VASTAI (BackendType): Vast.ai Marketplace
+ VERDA (BackendType): Verda Cloud
VULTR (BackendType): Vultr
"""
@@ -29,7 +30,7 @@ class BackendType(str, enum.Enum):
AZURE = "azure"
CLOUDRIFT = "cloudrift"
CUDO = "cudo"
- DATACRUNCH = "datacrunch"
+ DATACRUNCH = "datacrunch" # BackendType for backward compatibility
DIGITALOCEAN = "digitalocean"
DSTACK = "dstack"
GCP = "gcp"
@@ -43,4 +44,5 @@ class BackendType(str, enum.Enum):
RUNPOD = "runpod"
TENSORDOCK = "tensordock"
VASTAI = "vastai"
+ VERDA = "verda"
VULTR = "vultr"
diff --git a/src/dstack/_internal/server/testing/common.py b/src/dstack/_internal/server/testing/common.py
index 8f9459a766..0c4a1c5e7f 100644
--- a/src/dstack/_internal/server/testing/common.py
+++ b/src/dstack/_internal/server/testing/common.py
@@ -689,7 +689,7 @@ async def create_instance(
instance_id: Optional[UUID] = None,
job: Optional[JobModel] = None,
instance_num: int = 0,
- backend: BackendType = BackendType.DATACRUNCH,
+ backend: BackendType = BackendType.VERDA,
termination_policy: Optional[TerminationPolicy] = None,
termination_idle_time: int = DEFAULT_FLEET_TERMINATION_IDLE_TIME,
region: str = "eu-west",
diff --git a/src/tests/_internal/core/backends/cloudrift/test_configurator.py b/src/tests/_internal/core/backends/cloudrift/test_configurator.py
index f12499d890..89145279fd 100644
--- a/src/tests/_internal/core/backends/cloudrift/test_configurator.py
+++ b/src/tests/_internal/core/backends/cloudrift/test_configurator.py
@@ -12,7 +12,7 @@
from dstack._internal.core.errors import BackendInvalidCredentialsError
-class TestDataCrunchConfigurator:
+class TestCloudRiftConfigurator:
def test_validate_config_valid(self):
config = CloudRiftBackendConfigWithCreds(creds=CloudRiftCreds(api_key="valid"))
with patch(
diff --git a/src/tests/_internal/core/backends/datacrunch/test_configurator.py b/src/tests/_internal/core/backends/datacrunch/test_configurator.py
deleted file mode 100644
index 721cebe1a2..0000000000
--- a/src/tests/_internal/core/backends/datacrunch/test_configurator.py
+++ /dev/null
@@ -1,21 +0,0 @@
-from unittest.mock import patch
-
-from dstack._internal.core.backends.datacrunch.configurator import (
- DataCrunchConfigurator,
-)
-from dstack._internal.core.backends.datacrunch.models import (
- DataCrunchBackendConfigWithCreds,
- DataCrunchCreds,
-)
-
-
-class TestDataCrunchConfigurator:
- def test_validate_config_valid(self):
- config = DataCrunchBackendConfigWithCreds(
- creds=DataCrunchCreds(client_id="valid", client_secret="valid"),
- regions=["FIN-01"],
- )
- with patch(
- "dstack._internal.core.backends.datacrunch.configurator.DataCrunchConfigurator._validate_creds"
- ):
- DataCrunchConfigurator().validate_config(config, default_creds_enabled=True)
diff --git a/src/tests/_internal/core/backends/verda/__init__.py b/src/tests/_internal/core/backends/verda/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/src/tests/_internal/core/backends/verda/test_configurator.py b/src/tests/_internal/core/backends/verda/test_configurator.py
new file mode 100644
index 0000000000..a17fd1182c
--- /dev/null
+++ b/src/tests/_internal/core/backends/verda/test_configurator.py
@@ -0,0 +1,28 @@
+import sys
+from unittest.mock import patch
+
+import pytest
+
+if sys.version_info < (3, 10):
+ pytest.skip("Verda requires Python 3.10", allow_module_level=True)
+
+from dstack._internal.core.backends.verda.configurator import (
+ VerdaConfigurator,
+)
+from dstack._internal.core.backends.verda.models import (
+ VerdaBackendConfigWithCreds,
+ VerdaCreds,
+)
+
+
+class TestVerdaConfigurator:
+ def test_validate_config_valid(self):
+ config = VerdaBackendConfigWithCreds(
+ type="verda",
+ creds=VerdaCreds(client_id="valid", client_secret="valid"),
+ regions=["FIN-01"],
+ )
+ with patch(
+ "dstack._internal.core.backends.verda.configurator.VerdaConfigurator._validate_creds"
+ ):
+ VerdaConfigurator().validate_config(config, default_creds_enabled=True)
diff --git a/src/tests/_internal/server/background/tasks/test_process_instances.py b/src/tests/_internal/server/background/tasks/test_process_instances.py
index 8661b81ee0..e7c44ab434 100644
--- a/src/tests/_internal/server/background/tasks/test_process_instances.py
+++ b/src/tests/_internal/server/background/tasks/test_process_instances.py
@@ -553,7 +553,7 @@ class TestTerminate:
@contextmanager
def mock_terminate_in_backend(error: Optional[Exception] = None):
backend = Mock()
- backend.TYPE = BackendType.DATACRUNCH
+ backend.TYPE = BackendType.VERDA
terminate_instance = backend.compute.return_value.terminate_instance
if error is not None:
terminate_instance.side_effect = error
diff --git a/src/tests/_internal/server/routers/test_backends.py b/src/tests/_internal/server/routers/test_backends.py
index febe7d17e4..433c12de30 100644
--- a/src/tests/_internal/server/routers/test_backends.py
+++ b/src/tests/_internal/server/routers/test_backends.py
@@ -88,7 +88,7 @@ async def test_returns_backend_types(self, client: AsyncClient):
"azure",
"cloudrift",
"cudo",
- "datacrunch",
+ *(["datacrunch"] if sys.version_info >= (3, 10) else []),
"digitalocean",
"gcp",
"hotaisle",
@@ -98,6 +98,7 @@ async def test_returns_backend_types(self, client: AsyncClient):
"oci",
"runpod",
"vastai",
+ *(["verda"] if sys.version_info >= (3, 10) else []),
"vultr",
]
+ 