diff options
author | Bryan Newbold <bnewbold@robocracy.org> | 2021-11-04 17:08:36 -0700 |
---|---|---|
committer | Bryan Newbold <bnewbold@robocracy.org> | 2021-11-09 14:17:35 -0800 |
commit | 94ddfd7167994b4c0f7940317655d152aba302e6 (patch) | |
tree | 792097ca3e0df885f423476f9510bd69be1fc2eb /python | |
parent | c2cdf60d509e380029f6e2566fc4f98eff4b9f1a (diff) | |
download | fatcat-94ddfd7167994b4c0f7940317655d152aba302e6.tar.gz fatcat-94ddfd7167994b4c0f7940317655d152aba302e6.zip |
short wayback ts: initial cleanup script implementation
Diffstat (limited to 'python')
-rw-r--r-- | python/fatcat_tools/cleanups/file_short_wayback_ts.py | 251 |
1 files changed, 251 insertions, 0 deletions
diff --git a/python/fatcat_tools/cleanups/file_short_wayback_ts.py b/python/fatcat_tools/cleanups/file_short_wayback_ts.py new file mode 100644 index 00000000..2d893dbf --- /dev/null +++ b/python/fatcat_tools/cleanups/file_short_wayback_ts.py @@ -0,0 +1,251 @@ +import argparse +import copy +import os +import sys +from typing import Any, Dict + +import fatcat_openapi_client +from fatcat_openapi_client import ApiClient, FileEntity + +from fatcat_tools import authenticated_api, entity_from_dict, public_api +from fatcat_tools.importers.common import EntityImporter, JsonLinePusher + + +class FileShortWaybackTimestampCleanup(EntityImporter): + """ + This is a one-off / one-time cleanup script for file entities, fix short + timestamps in wayback URLs. These timestamps are supposed to have 14 digits + (datetime with year, hour, seconds, etc). Some legacy file imports ended up + with only 4 or 12 digits. + + While this calls itself a cleanup, it is based on the import code path. It + is not integrated into the `fatcat_import` or `fatcat_cleanup` controller; + instead it has a __main__ function and is invoked like: + + python -m fatcat_tools.cleans.file_short_wayback-ts < blah.json + """ + + def __init__(self, api: ApiClient, **kwargs): + + eg_desc = kwargs.pop("editgroup_description", None) or "Expand short wayback timestamps" + eg_extra = kwargs.pop("editgroup_extra", dict()) + eg_extra["agent"] = eg_extra.get( + "agent", "fatcat_tools.FileShortWaybackTimestampCleanup" + ) + super().__init__( + api, + do_updates=True, + editgroup_description=eg_desc, + editgroup_extra=eg_extra, + **kwargs + ) + self.testing_mode = False + + def want(self, row: Dict[str, Any]) -> bool: + if row["status"].startswith("success"): + return True + else: + self.counts["skip-status"] += 1 + return False + + def parse_record(self, row: Dict[str, Any]) -> FileEntity: + + # bezerk mode doesn't make sense for this importer + assert self.bezerk_mode is False + + fe: FileEntity = entity_from_dict(row["file_entity"], FileEntity) + status: str = row["status"] + assert status.startswith("success") + url_expansions: Dict[str, str] = row["full_urls"] + assert len(url_expansions) >= 1 + + # actual cleanup happens here + any_fixed = False + for fe_url in fe.urls: + if "://web.archive.org/web/" not in fe_url.url: + continue + seq = fe_url.url.split("/") + partial_ts = seq[4] + original_url = "/".join(seq[5:]) + if seq[2] != "web.archive.org": + continue + if len(partial_ts) not in [4, 12]: + continue + if fe_url.url in url_expansions: + fix_url = url_expansions[fe_url.url] + # defensive checks + assert partial_ts in fix_url + assert "://" in fix_url + assert fe_url.url.endswith(original_url) + assert fix_url.endswith(original_url) + fe_url.url = fix_url + any_fixed = True + + if not any_fixed: + self.counts["skip-no-fixes"] += 1 + return None + + # do any other generic file entity cleanups + # this includes removing duplicates + fe = self.generic_file_cleanups(fe) + + # verify that there are no exact duplicates + final_urls = [u.url for u in fe.urls] + assert len(final_urls) == len(list(set(final_urls))) + + return fe + + def try_update(self, fe: FileEntity) -> bool: + + # should always be existing + try: + existing = self.api.get_file(fe.ident) + except fatcat_openapi_client.rest.ApiException as err: + if err.status != 404: + raise err + + if not existing: + self.counts["skip-existing-not-found"] += 1 + return False + + if existing.sha1 != fe.sha1: + self.counts["skip-existing-mismatch"] += 1 + return False + + assert fe.revision and existing.revision + if existing.revision != fe.revision: + self.counts["skip-revision-changed"] += 1 + return False + + # verify that at least one URL remains + if not fe.urls or len(fe.urls) < 1: + self.counts["skip-no-urls"] += 1 + return False + + # verify that all wayback urls have 14-digit timestamps, and are generally well-formed + for u in fe.urls: + if "://web.archive.org/web/" not in u.url: + continue + if u.rel != "webarchive": + self.counts["skip-bad-wayback-rel"] += 1 + return False + seg = u.url.split("/") + if ( + len(seg) < 6 + or seg[0] != "https:" + or seg[2] != "web.archive.org" + or seg[3] != "web" + ): + self.counts["skip-bad-wayback"] += 1 + return False + if len(seg[4]) != 14 or not seg[4].isdigit(): + self.counts["skip-bad-wayback-timestamp"] += 1 + return False + + if existing == fe or existing.urls == fe.urls: + self.counts["skip-no-change"] += 1 + return False + + # not doing a check for "in current editgroup", because the source of + # these corrections (entity dump) contains no dupes + + if not self.testing_mode: + self.api.update_file(self.get_editgroup_id(), existing.ident, existing) + self.counts["update"] += 1 + return False + + +def test_short_wayback_ts() -> None: + api = public_api("http://localhost:9411/v0") + fswtc = FileShortWaybackTimestampCleanup(api=api) + fswtc.testing_mode = True + + assert fswtc.want({"status": "fail"}) is False + assert fswtc.want({"status": "success-self"}) is True + + example_line: Dict[str, Any] = { + "status": "success-db", + "file_entity": { + # note: doesn't match actual entity + "release_ids": ["waldfsctnbcpdbmasgduhaaaaa"], + "mimetype": "application/pdf", + "urls": [ + { + "url": "https://papiro.unizar.es/ojs/index.php/ais/article/download/2187/1971", + "rel": "web", + }, + { + "url": "https://web.archive.org/web/201904301022/https://papiro.unizar.es/ojs/index.php/ais/article/download/2187/1971", + "rel": "webarchive", + }, + ], + "sha256": "0b9e09480ed2e1f08f3c6c72f57ce12b52ea265f580f8810e606b49d64234b29", + "sha1": "be714299b9be21b5afdaa7affd7d710c58269433", + "md5": "9edb542be5b3446a1905e61a8a3abebd", + "size": 666242, + "revision": "fe949be8-7bf9-4c17-be28-8e3e90fb85bd", + "ident": "4ghpvs2t2rdtrdum2mkreh62me", + "state": "active", + }, + "full_urls": { + "https://web.archive.org/web/201904301022/https://papiro.unizar.es/ojs/index.php/ais/article/download/2187/1971": "https://web.archive.org/web/20190430102239/https://papiro.unizar.es/ojs/index.php/ais/article/download/2187/1971" + }, + } + example_fe = entity_from_dict(example_line["file_entity"], FileEntity) + + fe1 = copy.copy(example_fe) + fe1.urls[ + 1 + ].url = "https://web.archive.org/web/20190430102239/https://papiro.unizar.es/ojs/index.php/ais/article/download/2187/1971" + assert fswtc.parse_record(example_line) == fe1 + + # update code path; requires a known file ident and API running locally + assert fswtc.counts["update"] == 0 + dummy_fe = api.get_file("aaaaaaaaaaaaamztaaaaaaaaai") + fe1.ident = dummy_fe.ident + + assert fswtc.try_update(fe1) is False + assert fswtc.counts["skip-existing-mismatch"] == 1 + + fe1.sha1 = dummy_fe.sha1 + assert fswtc.try_update(fe1) is False + assert fswtc.counts["skip-revision-changed"] == 1 + + fe1.revision = dummy_fe.revision + assert fswtc.try_update(fe1) is False + print(fswtc.counts) + assert fswtc.counts["update"] == 1 + + +def main() -> None: + parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter) + parser.add_argument( + "--host-url", default="http://localhost:9411/v0", help="connect to this host/port" + ) + parser.add_argument("--batch-size", help="size of batch to send", default=50, type=int) + parser.set_defaults( + auth_var="FATCAT_API_AUTH_TOKEN", + ) + parser.add_argument( + "json_file", + help="File with jsonlines from file_meta schema to import from", + default=sys.stdin, + type=argparse.FileType("r"), + ) + + args = parser.parse_args() + api = authenticated_api( + args.host_url, + # token is an optional kwarg (can be empty string, None, etc) + token=os.environ.get(args.auth_var), + ) + + fswtc = FileShortWaybackTimestampCleanup( + api, + edit_batch_size=args.batch_size, + ) + JsonLinePusher(fswtc, args.json_file).run() + + +if __name__ == "__main__": + main() |