deltacat 1.1.1__tar.gz → 1.1.3__tar.gz

{deltacat-1.1.1 → deltacat-1.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deltacat
-Version: 1.1.1
+Version: 1.1.3
 Summary: A scalable, fast, ACID-compliant Data Catalog powered by Ray.
 Home-page: https://github.com/ray-project/deltacat
 Author: Ray Team

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/__init__.py

@@ -44,7 +44,7 @@ from deltacat.types.tables import TableWriteMode
 
 deltacat.logs.configure_deltacat_logger(logging.getLogger(__name__))
 
-__version__ = "1.1.1"
+__version__ = "1.1.3"
 
 
 __all__ = [

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/aws/clients.py

@@ -4,6 +4,7 @@ from typing import Optional
 from http import HTTPStatus
 
 import boto3
+from botocore.exceptions import CredentialRetrievalError
 from boto3.exceptions import ResourceNotExistsError
 from boto3.resources.base import ServiceResource
 from botocore.client import BaseClient
@@ -15,6 +16,8 @@ from tenacity import (
     wait_fixed,
     retry_if_exception,
     stop_after_delay,
+    retry_if_exception_type,
+    wait_random_exponential,
 )
 
 from deltacat import logs
@@ -37,6 +40,13 @@ RETRYABLE_HTTP_STATUS_CODES = [
     HTTPStatus.GATEWAY_TIMEOUT,
 ]
 
+boto_retry_wrapper = Retrying(
+    wait=wait_random_exponential(multiplier=1, max=10),
+    stop=stop_after_delay(60 * 5),
+    # CredentialRetrievalError can still be thrown due to throttling, even if IMDS health checks succeed.
+    retry=retry_if_exception_type(CredentialRetrievalError),
+)
+
 
 class RetryIfRetryableHTTPStatusCode(retry_if_exception):
     """
@@ -183,10 +193,10 @@ def _client(name: str, region: Optional[str], **kwargs) -> BaseClient:
 def resource_cache(name: str, region: Optional[str], **kwargs) -> ServiceResource:
     # we don't use the @lru_cache decorator because Ray can't pickle it
     cached_function = lru_cache()(_resource)
-    return cached_function(name, region, **kwargs)
+    return boto_retry_wrapper(cached_function, name, region, **kwargs)
 
 
 def client_cache(name: str, region: Optional[str], **kwargs) -> BaseClient:
     # we don't use the @lru_cache decorator because Ray can't pickle it
     cached_function = lru_cache()(_client)
-    return cached_function(name, region, **kwargs)
+    return boto_retry_wrapper(cached_function, name, region, **kwargs)

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/aws/constants.py

@@ -3,6 +3,10 @@ from typing import List
 from deltacat.utils.common import env_integer, env_string
 
 DAFT_MAX_S3_CONNECTIONS_PER_FILE = env_integer("DAFT_MAX_S3_CONNECTIONS_PER_FILE", 8)
-BOTO_MAX_RETRIES = env_integer("BOTO_MAX_RETRIES", 15)
+BOTO_MAX_RETRIES = env_integer("BOTO_MAX_RETRIES", 5)
 TIMEOUT_ERROR_CODES: List[str] = ["ReadTimeoutError", "ConnectTimeoutError"]
 AWS_REGION = env_string("AWS_REGION", "us-east-1")
+
+# Metric Names
+DOWNLOAD_MANIFEST_ENTRY_METRIC_PREFIX = "download_manifest_entry"
+UPLOAD_SLICED_TABLE_METRIC_PREFIX = "upload_sliced_table"

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/aws/s3u.py

@@ -25,7 +25,11 @@ from tenacity import (
 from deltacat.utils.ray_utils.concurrency import invoke_parallel
 import deltacat.aws.clients as aws_utils
 from deltacat import logs
-from deltacat.aws.constants import TIMEOUT_ERROR_CODES
+from deltacat.aws.constants import (
+    TIMEOUT_ERROR_CODES,
+    DOWNLOAD_MANIFEST_ENTRY_METRIC_PREFIX,
+    UPLOAD_SLICED_TABLE_METRIC_PREFIX,
+)
 from deltacat.exceptions import NonRetryableError, RetryableError
 from deltacat.storage import (
     DistributedDataset,
@@ -50,6 +54,7 @@ from deltacat.types.tables import (
 )
 from deltacat.types.partial_download import PartialFileDownloadParams
 from deltacat.utils.common import ReadKwargsProvider
+from deltacat.utils.metrics import metrics
 
 logger = logs.configure_deltacat_logger(logging.getLogger(__name__))
 
@@ -238,6 +243,7 @@ def read_file(
         raise e
 
 
+@metrics(prefix=UPLOAD_SLICED_TABLE_METRIC_PREFIX)
 def upload_sliced_table(
     table: Union[LocalTable, DistributedDataset],
     s3_url_prefix: str,
@@ -346,6 +352,7 @@ def upload_table(
     return manifest_entries
 
 
+@metrics(prefix=DOWNLOAD_MANIFEST_ENTRY_METRIC_PREFIX)
 def download_manifest_entry(
     manifest_entry: ManifestEntry,
     token_holder: Optional[Dict[str, Any]] = None,

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/constants.py

@@ -40,3 +40,31 @@ DROP_DUPLICATES = True
 # This is the observed upper bound inflation for parquet
 # size in metadata to pyarrow table size.
 PARQUET_TO_PYARROW_INFLATION = 4
+
+# Metric Names
+# Time taken for a hash bucket task
+HASH_BUCKET_TIME_IN_SECONDS = "hash_bucket_time"
+
+# Hash bucket success count
+HASH_BUCKET_SUCCESS_COUNT = "hash_bucket_success_count"
+
+# Hash bucket failure count
+HASH_BUCKET_FAILURE_COUNT = "hash_bucket_failure_count"
+
+# Time taken for a merge task
+MERGE_TIME_IN_SECONDS = "merge_time"
+
+# Merge success count
+MERGE_SUCCESS_COUNT = "merge_success_count"
+
+# Merge failure count
+MERGE_FAILURE_COUNT = "merge_failure_count"
+
+# Metric prefix for discover deltas
+DISCOVER_DELTAS_METRIC_PREFIX = "discover_deltas"
+
+# Metric prefix for prepare deletes
+PREPARE_DELETES_METRIC_PREFIX = "prepare_deletes"
+
+# Metric prefix for materialize
+MATERIALIZE_METRIC_PREFIX = "delta_materialize"

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/deletes/utils.py

@@ -23,6 +23,8 @@ from deltacat.storage import (
     Delta,
 )
 from deltacat import logs
+from deltacat.utils.metrics import metrics
+from deltacat.compute.compactor_v2.constants import PREPARE_DELETES_METRIC_PREFIX
 
 
 logger = logs.configure_deltacat_logger(logging.getLogger(__name__))
@@ -115,6 +117,7 @@ def _get_delete_file_envelopes(
     return delete_file_envelopes
 
 
+@metrics(prefix=PREPARE_DELETES_METRIC_PREFIX)
 def prepare_deletes(
     params: CompactPartitionParams,
     input_deltas: List[Delta],

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/steps/hash_bucket.py

@@ -25,12 +25,17 @@ from deltacat.utils.ray_utils.runtime import (
 )
 from deltacat.utils.common import ReadKwargsProvider
 from deltacat.utils.performance import timed_invocation
-from deltacat.utils.metrics import emit_timer_metrics
+from deltacat.utils.metrics import emit_timer_metrics, failure_metric, success_metric
 from deltacat.utils.resources import (
     get_current_process_peak_memory_usage_in_bytes,
     ProcessUtilizationOverTimeRange,
 )
 from deltacat.constants import BYTES_PER_GIBIBYTE
+from deltacat.compute.compactor_v2.constants import (
+    HASH_BUCKET_TIME_IN_SECONDS,
+    HASH_BUCKET_FAILURE_COUNT,
+    HASH_BUCKET_SUCCESS_COUNT,
+)
 
 if importlib.util.find_spec("memray"):
     import memray
@@ -91,6 +96,8 @@ def _group_file_records_by_pk_hash_bucket(
     return hb_to_delta_file_envelopes, total_record_count, total_size_bytes
 
 
+@success_metric(name=HASH_BUCKET_SUCCESS_COUNT)
+@failure_metric(name=HASH_BUCKET_FAILURE_COUNT)
 def _timed_hash_bucket(input: HashBucketInput):
     task_id = get_current_ray_task_id()
     worker_id = get_current_ray_worker_id()
@@ -153,7 +160,7 @@ def hash_bucket(input: HashBucketInput) -> HashBucketResult:
         if input.metrics_config:
             emit_result, latency = timed_invocation(
                 func=emit_timer_metrics,
-                metrics_name="hash_bucket",
+                metrics_name=HASH_BUCKET_TIME_IN_SECONDS,
                 value=duration,
                 metrics_config=input.metrics_config,
             )

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/steps/merge.py

@@ -24,7 +24,7 @@ from deltacat.utils.ray_utils.runtime import (
 )
 from deltacat.compute.compactor.utils import system_columns as sc
 from deltacat.utils.performance import timed_invocation
-from deltacat.utils.metrics import emit_timer_metrics
+from deltacat.utils.metrics import emit_timer_metrics, failure_metric, success_metric
 from deltacat.utils.resources import (
     get_current_process_peak_memory_usage_in_bytes,
     ProcessUtilizationOverTimeRange,
@@ -42,6 +42,11 @@ from deltacat.storage import (
 )
 from deltacat.compute.compactor_v2.utils.dedupe import drop_duplicates
 from deltacat.constants import BYTES_PER_GIBIBYTE
+from deltacat.compute.compactor_v2.constants import (
+    MERGE_TIME_IN_SECONDS,
+    MERGE_SUCCESS_COUNT,
+    MERGE_FAILURE_COUNT,
+)
 
 
 if importlib.util.find_spec("memray"):
@@ -479,6 +484,8 @@ def _copy_manifests_from_hash_bucketing(
     return materialized_results
 
 
+@success_metric(name=MERGE_SUCCESS_COUNT)
+@failure_metric(name=MERGE_FAILURE_COUNT)
 def _timed_merge(input: MergeInput) -> MergeResult:
     task_id = get_current_ray_task_id()
     worker_id = get_current_ray_worker_id()
@@ -578,7 +585,7 @@ def merge(input: MergeInput) -> MergeResult:
         if input.metrics_config:
             emit_result, latency = timed_invocation(
                 func=emit_timer_metrics,
-                metrics_name="merge",
+                metrics_name=MERGE_TIME_IN_SECONDS,
                 value=duration,
                 metrics_config=input.metrics_config,
             )

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/utils/io.py

@@ -23,10 +23,13 @@ from deltacat.compute.compactor_v2.utils.task_options import (
 from deltacat.compute.compactor_v2.utils.content_type_params import (
     append_content_type_params,
 )
+from deltacat.utils.metrics import metrics
+from deltacat.compute.compactor_v2.constants import DISCOVER_DELTAS_METRIC_PREFIX
 
 logger = logs.configure_deltacat_logger(logging.getLogger(__name__))
 
 
+@metrics(prefix=DISCOVER_DELTAS_METRIC_PREFIX)
 def discover_deltas(
     source_partition_locator: PartitionLocator,
     last_stream_position_to_compact: int,

{deltacat-1.1.1 → deltacat-1.1.3}/deltacat/compute/compactor_v2/utils/merge.py

@@ -31,11 +31,14 @@ from deltacat.compute.compactor_v2.deletes.delete_strategy import (
 from deltacat.compute.compactor_v2.deletes.delete_file_envelope import (
     DeleteFileEnvelope,
 )
+from deltacat.utils.metrics import metrics
+from deltacat.compute.compactor_v2.constants import MATERIALIZE_METRIC_PREFIX
 
 
 logger = logs.configure_deltacat_logger(logging.getLogger(__name__))
 
 
+@metrics(prefix=MATERIALIZE_METRIC_PREFIX)
 def materialize(
     input: MergeInput,
     task_index: int,