|
| 1 | +import logging |
| 2 | +import pathlib |
| 3 | +import traceback |
| 4 | +from typing import Any |
| 5 | + |
| 6 | +import numpy as np |
| 7 | + |
| 8 | +from .meta_cache_sqlite import SQLiteKeyJSONDatabase |
| 9 | + |
| 10 | + |
| 11 | +logger = logging.getLogger(__name__) |
| 12 | + |
| 13 | + |
| 14 | +class MetaCache: |
| 15 | + """Cache dataset dictionaries (metadata) |
| 16 | +
|
| 17 | + The implementation uses an SQLite database which is loaded upon |
| 18 | + init and edited whenever data changes. For compute-intensive |
| 19 | + tasks (searching), the metadata are loaded into memory. |
| 20 | + """ |
| 21 | + def __init__(self, |
| 22 | + directory: str | pathlib.Path, |
| 23 | + circle_ids: list[str] = None, |
| 24 | + ) -> None: |
| 25 | + """ |
| 26 | + Scan *directory* for ``circle_*.db`` files, load all of them |
| 27 | + and fill the numpy backing store. |
| 28 | +
|
| 29 | + Parameters |
| 30 | + ---------- |
| 31 | + directory : str | pathlib.Path |
| 32 | + Path to the folder that will hold the ``circle_<org_id>.db`` files. |
| 33 | + The folder is created automatically if it does not exist. |
| 34 | + circle_ids : list[str] |
| 35 | + List of circle IDs that should be taken into consideration. |
| 36 | + If set to None (default), all databases in the `directory` |
| 37 | + are loaded. |
| 38 | + """ |
| 39 | + self.base_dir = pathlib.Path(directory).expanduser().resolve() |
| 40 | + self.base_dir.mkdir(parents=True, exist_ok=True) |
| 41 | + |
| 42 | + # The registry is a dictionary with circle IDs and a list of |
| 43 | + # dataset IDs as values. |
| 44 | + self._registry_org = {} |
| 45 | + |
| 46 | + # List of all dataset dictionaries (only used during init) |
| 47 | + datasets = [] |
| 48 | + # List of search data (only used during init) |
| 49 | + rows: list[tuple] = [] |
| 50 | + |
| 51 | + # Initialize the databases |
| 52 | + if circle_ids is None: |
| 53 | + db_paths = list(self.base_dir.glob("circle_*.db")) |
| 54 | + else: |
| 55 | + db_paths = [self.base_dir / f"circle_{c}.db" for c in circle_ids] |
| 56 | + # Dictionary of databases for persistent storage |
| 57 | + self._databases = {} |
| 58 | + for cp in db_paths: |
| 59 | + cid = cp.stem.split("_", 1)[-1] |
| 60 | + try: |
| 61 | + self._databases[cid] = SQLiteKeyJSONDatabase(cp) |
| 62 | + except BaseException: |
| 63 | + logger.error( |
| 64 | + f"Recreating broken DB '{cp}': {traceback.format_exc()}") |
| 65 | + cp.unlink() |
| 66 | + self._databases[cid] = SQLiteKeyJSONDatabase(cp) |
| 67 | + |
| 68 | + # populate registry, dataset list, and search array data |
| 69 | + # initial blob size for dataset search |
| 70 | + blob_size = 256 |
| 71 | + for db in self._databases.values(): |
| 72 | + for ds_dict in db: |
| 73 | + ds_id: str = ds_dict.get("id", "") |
| 74 | + m_created: str = ds_dict.get("metadata_created", "") |
| 75 | + |
| 76 | + # Build a double-space‑separated string of **only values** |
| 77 | + value_blob = _create_blob_for_search(ds_dict) |
| 78 | + blob_size = max(blob_size, len(value_blob)) |
| 79 | + |
| 80 | + rows.append((ds_id, m_created, value_blob)) |
| 81 | + datasets.append(ds_dict) |
| 82 | + self._registry_org.setdefault(ds_dict["owner_org"], |
| 83 | + []).append(ds_dict["id"]) |
| 84 | + |
| 85 | + # Convert the Python list of tuples into a NumPy 2‑D object array |
| 86 | + # TODO: If there are performance issues, we can create these arrays |
| 87 | + # in chunks of e.g. 10000 in the above for-loop. |
| 88 | + data_dtype = [ |
| 89 | + # "4b1f7c53-9d7f-2c53-aeb6-eaa8ecf10ca9" |
| 90 | + ("id", "<U36"), |
| 91 | + # 2025-09-18T08:09:52.947634 |
| 92 | + ("created", "<U26"), |
| 93 | + # initialize with maximum size |
| 94 | + ("blob", f"<U{blob_size}") |
| 95 | + ] |
| 96 | + if rows: |
| 97 | + data = np.array(rows, dtype=data_dtype) # shape (n_records, 3) |
| 98 | + else: |
| 99 | + # initialize emtpy cache |
| 100 | + data = np.empty(0, dtype=data_dtype) |
| 101 | + |
| 102 | + # Sort the dataset according to creation date, descending |
| 103 | + sort_idx = np.argsort(data["created"])[::-1] |
| 104 | + #: Blobs for searching, sorted by creation date descending |
| 105 | + self._srt_blobs = data[sort_idx] |
| 106 | + |
| 107 | + #: list of datasets, sorted by creation date descending |
| 108 | + self.datasets = [datasets[ii] for ii in sort_idx] |
| 109 | + |
| 110 | + def __enter__(self): |
| 111 | + return self |
| 112 | + |
| 113 | + def __exit__(self, exc_type, exc_val, exc_tb): |
| 114 | + self.close() |
| 115 | + |
| 116 | + def close(self): |
| 117 | + for db in self._databases.values(): |
| 118 | + db.close() |
| 119 | + |
| 120 | + def search(self, |
| 121 | + query: str, |
| 122 | + limit: int = None |
| 123 | + ) -> list[dict[str, Any]]: |
| 124 | + """ |
| 125 | + Free‑text search across **all** cached datasets. |
| 126 | +
|
| 127 | + The search is case‑insensitive and matches if the query appears |
| 128 | + anywhere in the JSON blob created from the database. |
| 129 | +
|
| 130 | + Parameters |
| 131 | + ---------- |
| 132 | + query: str |
| 133 | + Text to look for. |
| 134 | + limit: int | None, optional |
| 135 | + Maximum number of results to return. ``None`` (default) returns all |
| 136 | + matches. |
| 137 | +
|
| 138 | + Returns |
| 139 | + ------- |
| 140 | + List[dict] |
| 141 | + A list of the matching dataset dictionaries |
| 142 | + """ |
| 143 | + norm_query = query.strip().lower() |
| 144 | + if not norm_query: |
| 145 | + return [] # empty query -> no results |
| 146 | + |
| 147 | + # `np.strings.find` works element‑wise on an array of strings |
| 148 | + # and returns the index of the first occurrence (or -1 if not found). |
| 149 | + # `self._srt_blobs[:, 2]` is the column that holds the lower‑cased |
| 150 | + # blobs. |
| 151 | + match_mask = np.strings.find(self._srt_blobs["blob"], norm_query) != -1 |
| 152 | + |
| 153 | + if limit is not None: |
| 154 | + # Remove all items that are above a threshold. |
| 155 | + # TODO: apply the limit already during search by chunking? |
| 156 | + idx_lim = np.where(np.cumsum(match_mask) > limit)[0] |
| 157 | + if idx_lim.size: |
| 158 | + match_mask[idx_lim.min():] = False |
| 159 | + |
| 160 | + idx_result = np.where(match_mask)[0] |
| 161 | + |
| 162 | + return [self.datasets[idx] for idx in idx_result] |
| 163 | + |
| 164 | + def upsert_dataset(self, ds_dict: dict[str, Any]) -> None: |
| 165 | + """Insert a new dataset or update an existing one |
| 166 | +
|
| 167 | + If the dataset `id` is not present in the cache, it is |
| 168 | + appended to the persistent SQLite database and to the in‑memory |
| 169 | + structures. |
| 170 | +
|
| 171 | + If a dataset with the same `id` already exists, the record is |
| 172 | + replaced everywhere. |
| 173 | +
|
| 174 | + Parameters |
| 175 | + ---------- |
| 176 | + ds_dict : dict |
| 177 | + A CKAN‑style dataset dictionary. |
| 178 | + """ |
| 179 | + ds_id = ds_dict.get("id") |
| 180 | + org_id = ds_dict.get("owner_org") |
| 181 | + |
| 182 | + # Is this dataset new? |
| 183 | + if ds_id not in self._registry_org.setdefault(org_id, []): |
| 184 | + # We have a new dataset |
| 185 | + self._upsert_dataset_insert(ds_dict) |
| 186 | + else: |
| 187 | + # We have an existing dataset |
| 188 | + if self._databases[org_id][ds_id] != ds_dict: |
| 189 | + # We have tp update the dataset |
| 190 | + self._upsert_dataset_update(ds_dict) |
| 191 | + |
| 192 | + def _upsert_dataset_insert(self, ds_dict): |
| 193 | + """Insert a new dataset |
| 194 | +
|
| 195 | + The search array will be rebuilt fully, because the order of |
| 196 | + datasets will change. |
| 197 | + """ |
| 198 | + ds_id = ds_dict["id"] |
| 199 | + org_id = ds_dict["owner_org"] |
| 200 | + m_created = ds_dict["metadata_created"] |
| 201 | + blob = _create_blob_for_search(ds_dict) |
| 202 | + if len(blob) > int(self._srt_blobs.dtype["blob"].str[2:]): |
| 203 | + # Increase the search blob size. |
| 204 | + new_dtype = [("id", "<U36"), |
| 205 | + ("created", "<U26"), |
| 206 | + ("blob", f"<U{len(blob) + 10}") |
| 207 | + ] |
| 208 | + else: |
| 209 | + new_dtype = self._srt_blobs.dtype |
| 210 | + |
| 211 | + # registry |
| 212 | + self._registry_org.setdefault(org_id, []).append(ds_id) |
| 213 | + |
| 214 | + # search array |
| 215 | + dates = np.array(self._srt_blobs["created"]) |
| 216 | + new_size = dates.size + 1 |
| 217 | + new_idx = np.searchsorted(dates, m_created) |
| 218 | + dates.resize(new_size) |
| 219 | + new_blobs = np.empty(new_size, dtype=new_dtype) |
| 220 | + new_blobs[:new_idx] = self._srt_blobs[:new_idx] |
| 221 | + new_blobs[new_idx] = (ds_id, m_created, blob) |
| 222 | + new_blobs[new_idx + 1:] = self._srt_blobs[new_idx:] |
| 223 | + self._srt_blobs = new_blobs |
| 224 | + |
| 225 | + # datasets |
| 226 | + self.datasets.insert(new_idx, ds_dict) |
| 227 | + |
| 228 | + # persistent database |
| 229 | + if org_id not in self._databases: |
| 230 | + self._databases[org_id] = SQLiteKeyJSONDatabase( |
| 231 | + db_name=self.base_dir / f"circle_{org_id}.db") |
| 232 | + self._databases[org_id][ds_id] = ds_dict |
| 233 | + |
| 234 | + def _upsert_dataset_update(self, ds_dict): |
| 235 | + """Update an existing dataset |
| 236 | +
|
| 237 | + If the new blob for `ds_dict` is larger than the existing, |
| 238 | + a new copy of the search array is created in memory. |
| 239 | + """ |
| 240 | + ds_id = ds_dict["id"] |
| 241 | + org_id = ds_dict["owner_org"] |
| 242 | + |
| 243 | + # registry does not need updating (only contains ds_id) |
| 244 | + |
| 245 | + # Find the index in the database |
| 246 | + idx = np.where(self._srt_blobs["id"] == ds_id)[0][0] |
| 247 | + |
| 248 | + # search array |
| 249 | + blob = _create_blob_for_search(ds_dict) |
| 250 | + if len(blob) > int(self._srt_blobs.dtype["blob"].str[2:]): |
| 251 | + # Rewrite the search blobs, because this blob is bigger than any |
| 252 | + # of the blobs before. |
| 253 | + new_dtype = [("id", "<U36"), |
| 254 | + ("created", "<U26"), |
| 255 | + ("blob", f"<U{len(blob) + 10}") |
| 256 | + ] |
| 257 | + self._srt_blobs = np.array(self._srt_blobs, dtype=new_dtype) |
| 258 | + self._srt_blobs["blob"][idx] = blob |
| 259 | + |
| 260 | + # cached datasets |
| 261 | + self.datasets[idx] = ds_dict |
| 262 | + |
| 263 | + # persistent database |
| 264 | + self._databases[org_id][ds_id] = ds_dict |
| 265 | + |
| 266 | + |
| 267 | +def _create_blob_for_search(ds_dict): |
| 268 | + """Create a string blob from a dataset dictionary for free text search""" |
| 269 | + values = _values_only(ds_dict, |
| 270 | + only_keys=[ |
| 271 | + # dataset |
| 272 | + "authors", |
| 273 | + "creator_user_id", |
| 274 | + "doi", |
| 275 | + "id", |
| 276 | + "name", |
| 277 | + "notes", |
| 278 | + "title", |
| 279 | + # resource |
| 280 | + "resources", |
| 281 | + "dc:experiment:date", |
| 282 | + "dc:experiment:sample", |
| 283 | + "dc:setup:chip region", |
| 284 | + "dc:setup:identifier", |
| 285 | + "dc:setup:module composition", |
| 286 | + "description", |
| 287 | + # "id", # duplicate |
| 288 | + # "name", # duplicate |
| 289 | + "organization", |
| 290 | + # tags |
| 291 | + "tags", |
| 292 | + "display_name", |
| 293 | + # groups |
| 294 | + "groups", |
| 295 | + ]) |
| 296 | + value_blob: str = " ".join(values).lower() |
| 297 | + return value_blob |
| 298 | + |
| 299 | + |
| 300 | +def _values_only(obj: Any, |
| 301 | + only_keys: list[str], |
| 302 | + ) -> list[str]: |
| 303 | + """ |
| 304 | + Recursively walk a JSON‑compatible object and return a flat list of the |
| 305 | + string representation of all values whose keys are in `only_keys`. |
| 306 | + Lists, tuples and dicts are traversed; other scalar types are converted |
| 307 | + with ``str``. |
| 308 | + """ |
| 309 | + vals = [] |
| 310 | + |
| 311 | + if isinstance(obj, dict): |
| 312 | + for k, v in obj.items(): |
| 313 | + if k in only_keys: |
| 314 | + vals += _values_only(v, only_keys) |
| 315 | + elif isinstance(obj, (list, tuple)): |
| 316 | + for v in obj: |
| 317 | + vals += _values_only(v, only_keys) |
| 318 | + else: |
| 319 | + # scalar (str, int, float, bool, None) |
| 320 | + if obj not in [None, ""]: |
| 321 | + vals.append(str(obj)) |
| 322 | + return vals |
0 commit comments