shopdb-flask/scripts/collapse_duplicate_backup_revisions.py

"""Collapse the duplicate backup revisions left by the shared-machine-number bug.

Dedup used to compare a posted config against the latest revision for the
ASSET. That is correct only when a machine number means one PC. Several PCs
legitimately share one here - the part markers on 0613, 0615 and WJPRT, whose
configs differ by COM port - so each marker's post differed from whichever
marker had posted last, nothing ever deduped, and the table grew by one row per
PC per collection cycle. Dedup is now per source hostname; this cleans up what
the old rule wrote.

A machine number that is NOT meant to be shared but shows two sources here is a
different fault - two PCs carrying the same machine number - and this script
will report it as two chains. Deal with that at the PC, not in the database.

WHAT IT REMOVES, and nothing else: within one chain (asset + kind + source
hostname), a revision whose content hash equals the revision immediately before
it from the same source. Those rows record no change and are exactly the rows
the fixed code would not have created.

WHAT IT KEEPS: every genuine change, in order; the newest revision of every
chain; the oldest revision of every chain; and every chain of every source, so
a quiet PC's only backup is never touched. A config that reverts to an earlier
state keeps both rows, because that is a real change and not a duplicate.

Share-backed kinds (partmarker) are metadata rows pointing at files on the
share. Deleting a row does not touch the file.

Usage (on the target instance, in the app dir):
    venv\\Scripts\\python scripts\\collapse_duplicate_backup_revisions.py
    venv\\Scripts\\python scripts\\collapse_duplicate_backup_revisions.py --commit
    ... --assetnumber 0615        # limit to one machine
    ... --kind ntlars             # limit to one kind

Dry run by default: it prints what it would delete and changes nothing. Take a
database backup before --commit anyway; this deletes rows.
"""

import argparse
import os
import sys
from pathlib import Path
from collections import defaultdict

# Run from anywhere: put the repo root (parent of scripts/) on the path.
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from dotenv import load_dotenv

# Load .env before anything reads config. wsgi.py does this for the served app
# and the `flask` CLI does it for itself, but a plain `python scripts/...` run
# gets neither: DATABASE_URL stays unset, FLASK_ENV defaults to development,
# and config falls back to a hardcoded root@localhost URL. That fails on a real
# deployment with a bare 1045 Access denied for 'root', naming a user the site
# does not use - and on a box where those root credentials happen to work, it
# is worse, because the script silently reads or WRITES the wrong database.
load_dotenv(Path(__file__).resolve().parent.parent / '.env')


from shopdb import create_app
from shopdb.api import db
from shopdb.core.models import Asset
from plugins.backups.models import BackupRevision


def _chains(assetnumber=None, kind=None):
    """Every revision grouped into (assetid, kind, sourcehostname) chains."""
    query = db.session.query(BackupRevision)
    if assetnumber:
        asset = db.session.query(Asset).filter(
            Asset.assetnumber == assetnumber).first()
        if asset is None:
            raise SystemExit('no asset with assetnumber {!r}'.format(assetnumber))
        query = query.filter(BackupRevision.assetid == asset.assetid)
    if kind:
        query = query.filter(BackupRevision.backupkind == kind)

    chains = defaultdict(list)
    for revision in query.order_by(BackupRevision.backuprevisionid.asc()).all():
        key = (revision.assetid, revision.backupkind, revision.sourcehostname)
        chains[key].append(revision)
    return chains


def _redundant(revisions):
    """Revisions in one chain that repeat the hash of the one before them.

    The chain is in id order, so this walks it forwards and keeps the first
    appearance of each run. A later run of the same hash after a genuine change
    is kept: that is a revert, which is a real event.
    """
    doomed = []
    previoushash = None
    for revision in revisions:
        if previoushash is not None and revision.contenthash == previoushash:
            doomed.append(revision)
        else:
            previoushash = revision.contenthash
    # Never drop the newest row of a chain, even if it repeats: it is the row
    # the asset panel and a restore both read.
    newestid = revisions[-1].backuprevisionid
    return [r for r in doomed if r.backuprevisionid != newestid]


def _report(chains):
    """Say what grew each chain, so a real cause can be told from the shared
    machine numbers.

    Three findings, and they need different fixes:

      several sources on one machine number - either a legitimately shared
      number (0613, 0615, WJPRT) or, on any other number, two PCs carrying the
      same machine number, which is a fault to fix on the PC.

      one source, many revisions, few distinct hashes - the old asset-wide
      dedup, or a value flapping between two states.

      one source, many revisions, all distinct hashes - something in the
      captured config changes on its own. The value names listed are the ones
      that changed most often between consecutive revisions; a timestamp or a
      session counter in the projection would show up here, and the fix is to
      exclude it from the hash, not to delete rows.
    """
    from plugins.backups.api.routes import _diffprojections

    bymachine = defaultdict(list)
    for key, revisions in chains.items():
        bymachine[key[0], key[1]].append((key[2], revisions))

    for (assetid, kind), sources in sorted(bymachine.items()):
        asset = db.session.get(Asset, assetid)
        name = asset.assetnumber if asset else str(assetid)
        total = sum(len(r) for _, r in sources)
        print('{} / {}: {} revisions from {} source(s)'.format(
            name, kind, total, len(sources)))

        for source, revisions in sorted(sources, key=lambda s: s[0] or ''):
            hashes = {r.contenthash for r in revisions}
            print('    {:<20} {:>5} revisions, {:>4} distinct'.format(
                source or '(no source)', len(revisions), len(hashes)))

            if len(revisions) < 3 or kind != 'ntlars':
                continue
            churn = defaultdict(int)
            for older, newer in zip(revisions, revisions[1:]):
                if older.contenthash == newer.contenthash:
                    continue
                for change in _diffprojections(older.payload, newer.payload):
                    churn['{}\\{}'.format(change['keypath'],
                                          change['valuename'])] += 1
            for valuename, count in sorted(churn.items(),
                                           key=lambda kv: -kv[1])[:5]:
                print('        changed {:>4}x  {}'.format(count, valuename))
        print()


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument('--commit', action='store_true',
                        help='apply the deletions (default is a dry run)')
    parser.add_argument('--report', action='store_true',
                        help='explain what grew each chain and delete nothing')
    parser.add_argument('--assetnumber', help='limit to one machine number')
    parser.add_argument('--kind', help='limit to one backup kind')
    args = parser.parse_args()

    app = create_app(os.environ.get('FLASK_ENV', 'production'))
    with app.app_context():
        chains = _chains(args.assetnumber, args.kind)
        if not chains:
            print('No revisions matched.')
            return

        if args.report:
            _report(chains)
            return

        assetnames = {}
        total = 0
        machines = defaultdict(int)

        for (assetid, kind, source), revisions in sorted(chains.items(),
                                                         key=lambda kv: kv[0]):
            doomed = _redundant(revisions)
            if not doomed:
                continue
            if assetid not in assetnames:
                asset = db.session.get(Asset, assetid)
                assetnames[assetid] = (asset.assetnumber if asset
                                       else str(assetid))
            name = assetnames[assetid]
            machines[name] += len(doomed)
            total += len(doomed)
            print('{:>10}  {:<12} {:<20} {} of {} redundant'.format(
                name, kind, source or '(no source)', len(doomed),
                len(revisions)))
            for revision in doomed:
                db.session.delete(revision)

        print()
        for name in sorted(machines):
            print('{:>10}: {} rows'.format(name, machines[name]))
        print('{} chains, {} redundant revisions'.format(len(chains), total))

        if not total:
            print('Nothing to do.')
            return
        if args.commit:
            db.session.commit()
            print('DELETED {} revisions.'.format(total))
        else:
            db.session.rollback()
            print('Dry run - nothing changed. Re-run with --commit to apply.')


if __name__ == '__main__':
    main()