|
| 1 | +"""Retry-wrapped managed-database client shared by Hotdata adapter packages. |
| 2 | +
|
| 3 | +Both hotdata-airflow and hotdata-dlt-destination import this module so that |
| 4 | +the higher-level client logic (retries, Arrow queries, table management) lives |
| 5 | +in one place rather than being duplicated per adapter. |
| 6 | +""" |
| 7 | + |
| 8 | +from __future__ import annotations |
| 9 | + |
| 10 | +import time |
| 11 | +from collections.abc import Callable |
| 12 | +from typing import Any, TypeVar |
| 13 | + |
| 14 | +import pyarrow as pa |
| 15 | +from hotdata.api.query_api import QueryApi |
| 16 | +from hotdata.api.query_runs_api import QueryRunsApi |
| 17 | +from hotdata.api.results_api import ResultsApi |
| 18 | +from hotdata.arrow import ResultsApi as ArrowResultsApi |
| 19 | +from hotdata.models.async_query_response import AsyncQueryResponse |
| 20 | +from hotdata.models.query_request import QueryRequest |
| 21 | +from hotdata.models.query_response import QueryResponse |
| 22 | +from hotdata_runtime.client import HotdataClient as RuntimeClient |
| 23 | +from hotdata_runtime.databases import LoadManagedTableResult, ManagedDatabase |
| 24 | +from hotdata_runtime.errors import ( |
| 25 | + HotdataTerminalError, |
| 26 | + HotdataTransientError, |
| 27 | + classify_sdk_error, |
| 28 | +) |
| 29 | + |
| 30 | +T = TypeVar("T") |
| 31 | + |
| 32 | + |
| 33 | +class ManagedDatabaseClient: |
| 34 | + """Managed-database client with bounded retries over hotdata-runtime. |
| 35 | +
|
| 36 | + This is the shared client used by Hotdata adapter packages (Airflow, |
| 37 | + dlt, etc.). It wraps the lower-level RuntimeClient with retry logic, |
| 38 | + Arrow-based result fetching, and convenience helpers for the managed |
| 39 | + database lifecycle. |
| 40 | + """ |
| 41 | + |
| 42 | + def __init__( |
| 43 | + self, |
| 44 | + *, |
| 45 | + api_key: str, |
| 46 | + workspace_id: str, |
| 47 | + api_base_url: str, |
| 48 | + max_retries: int, |
| 49 | + retry_backoff_seconds: float, |
| 50 | + ) -> None: |
| 51 | + self._max_retries = max_retries |
| 52 | + self._retry_backoff_seconds = retry_backoff_seconds |
| 53 | + self._runtime = RuntimeClient( |
| 54 | + api_key, |
| 55 | + workspace_id, |
| 56 | + host=api_base_url.rstrip("/"), |
| 57 | + ) |
| 58 | + |
| 59 | + def close(self) -> None: |
| 60 | + self._runtime.close() |
| 61 | + |
| 62 | + def ensure_managed_database( |
| 63 | + self, |
| 64 | + name: str, |
| 65 | + *, |
| 66 | + schema: str, |
| 67 | + tables: list[str], |
| 68 | + create_if_missing: bool, |
| 69 | + ) -> ManagedDatabase: |
| 70 | + def operation() -> ManagedDatabase: |
| 71 | + try: |
| 72 | + return self._runtime.resolve_managed_database(name) |
| 73 | + except KeyError: |
| 74 | + if not create_if_missing: |
| 75 | + raise |
| 76 | + return self._runtime.create_managed_database( |
| 77 | + description=name, |
| 78 | + schema=schema, |
| 79 | + tables=sorted(set(tables)), |
| 80 | + ) |
| 81 | + |
| 82 | + return self._request_with_retry(operation) |
| 83 | + |
| 84 | + def table_is_synced(self, database: str, table: str, *, schema: str) -> bool: |
| 85 | + for managed_table in self._runtime.list_managed_tables(database, schema=schema): |
| 86 | + if managed_table.table == table: |
| 87 | + return managed_table.synced |
| 88 | + return False |
| 89 | + |
| 90 | + def fetch_table(self, *, database: str, schema: str, table: str) -> pa.Table | None: |
| 91 | + def operation() -> pa.Table | None: |
| 92 | + if not self.table_is_synced(database, table, schema=schema): |
| 93 | + return None |
| 94 | + db = self._runtime.resolve_managed_database(database) |
| 95 | + sql = f'SELECT * FROM "default"."{schema}"."{table}"' |
| 96 | + result_id = self._query_database_scoped(sql, database_id=db.id) |
| 97 | + if result_id is None: |
| 98 | + return None |
| 99 | + return ArrowResultsApi(self._runtime.api).get_result_arrow(result_id) |
| 100 | + |
| 101 | + return self._request_with_retry(operation) |
| 102 | + |
| 103 | + _QUERY_TIMEOUT_SECONDS = 300.0 |
| 104 | + |
| 105 | + def _query_database_scoped(self, sql: str, *, database_id: str) -> str | None: |
| 106 | + raw = QueryApi(self._runtime.api).query( |
| 107 | + QueryRequest(sql=sql), |
| 108 | + x_database_id=database_id, |
| 109 | + ) |
| 110 | + if isinstance(raw, QueryResponse): |
| 111 | + return raw.result_id |
| 112 | + |
| 113 | + if isinstance(raw, AsyncQueryResponse): |
| 114 | + runs = QueryRunsApi(self._runtime.api) |
| 115 | + deadline = time.monotonic() + self._QUERY_TIMEOUT_SECONDS |
| 116 | + result_id: str | None = None |
| 117 | + while time.monotonic() < deadline: |
| 118 | + run = runs.get_query_run(raw.query_run_id) |
| 119 | + if run.status == "succeeded": |
| 120 | + result_id = run.result_id |
| 121 | + break |
| 122 | + if run.status in ("failed", "cancelled"): |
| 123 | + raise RuntimeError(run.error_message or f"Query {run.status}") |
| 124 | + time.sleep(0.5) |
| 125 | + else: |
| 126 | + raise TimeoutError( |
| 127 | + f"Managed database query timed out after {self._QUERY_TIMEOUT_SECONDS}s" |
| 128 | + ) |
| 129 | + return self._wait_result_ready(result_id) |
| 130 | + |
| 131 | + return None |
| 132 | + |
| 133 | + def _wait_result_ready(self, result_id: str | None) -> str | None: |
| 134 | + if result_id is None: |
| 135 | + return None |
| 136 | + results = ResultsApi(self._runtime.api) |
| 137 | + deadline = time.monotonic() + self._QUERY_TIMEOUT_SECONDS |
| 138 | + while time.monotonic() < deadline: |
| 139 | + r = results.get_result(result_id) |
| 140 | + if r.status == "ready": |
| 141 | + return result_id |
| 142 | + if r.status in ("failed", "cancelled"): |
| 143 | + raise RuntimeError(r.error_message or f"Result {r.status}") |
| 144 | + time.sleep(0.3) |
| 145 | + raise TimeoutError(f"Result {result_id} not ready after {self._QUERY_TIMEOUT_SECONDS}s") |
| 146 | + |
| 147 | + def fetch_table_rows(self, *, database: str, schema: str, table: str) -> list[dict[str, Any]]: |
| 148 | + result = self.fetch_table(database=database, schema=schema, table=table) |
| 149 | + return result.to_pylist() if result is not None else [] |
| 150 | + |
| 151 | + def upload_parquet(self, path: str) -> str: |
| 152 | + return self._request_with_retry(lambda: self._runtime.upload_parquet(path)) |
| 153 | + |
| 154 | + def load_managed_table( |
| 155 | + self, |
| 156 | + database: str, |
| 157 | + table: str, |
| 158 | + *, |
| 159 | + schema: str, |
| 160 | + upload_id: str, |
| 161 | + ) -> LoadManagedTableResult: |
| 162 | + return self._request_with_retry( |
| 163 | + lambda: self._runtime.load_managed_table( |
| 164 | + database, |
| 165 | + table, |
| 166 | + schema=schema, |
| 167 | + upload_id=upload_id, |
| 168 | + ) |
| 169 | + ) |
| 170 | + |
| 171 | + _MAX_BACKOFF_SECONDS = 30.0 |
| 172 | + |
| 173 | + def _request_with_retry(self, operation: Callable[[], T]) -> T: |
| 174 | + for attempt in range(1, self._max_retries + 1): |
| 175 | + try: |
| 176 | + return operation() |
| 177 | + except Exception as error: |
| 178 | + mapped_error = classify_sdk_error(error.__cause__ or error) |
| 179 | + if isinstance(mapped_error, HotdataTransientError) and attempt < self._max_retries: |
| 180 | + backoff = min(self._retry_backoff_seconds * attempt, self._MAX_BACKOFF_SECONDS) |
| 181 | + time.sleep(backoff) |
| 182 | + continue |
| 183 | + raise mapped_error from error |
| 184 | + raise RuntimeError("No retry attempts configured") |
0 commit comments