lbry-sdk/lbrynet/db_migrator/migrate4to5.py

64 lines
2.3 KiB
Python
Raw Normal View History

import sqlite3
import os
import logging
log = logging.getLogger(__name__)
def do_migration(db_dir):
log.info("Doing the migration")
add_lbry_file_metadata(db_dir)
log.info("Migration succeeded")
def add_lbry_file_metadata(db_dir):
"""
We migrate the blobs.db used in BlobManager to have a "should_announce" column,
and set this to True for blobs that are sd_hash's or head blobs (first blob in stream)
"""
name_metadata = os.path.join(db_dir, "blockchainname.db")
lbryfile_info_db = os.path.join(db_dir, 'lbryfile_info.db')
if not os.path.isfile(name_metadata) and not os.path.isfile(lbryfile_info_db):
return
if not os.path.isfile(lbryfile_info_db):
log.error(
"blockchainname.db was not found but lbryfile_info.db was found, skipping migration")
return
name_metadata_db = sqlite3.connect(name_metadata)
lbryfile_db = sqlite3.connect(lbryfile_info_db)
name_metadata_cursor = name_metadata_db.cursor()
lbryfile_cursor = lbryfile_db.cursor()
lbryfile_db.executescript(
"create table if not exists lbry_file_metadata (" +
" lbry_file integer primary key, " +
" txid text, " +
" n integer, " +
" foreign key(lbry_file) references lbry_files(rowid)"
")")
_files = lbryfile_cursor.execute("select rowid, stream_hash from lbry_files").fetchall()
lbry_files = {x[1]: x[0] for x in _files}
for (sd_hash, stream_hash) in lbryfile_cursor.execute("select * "
"from lbry_file_descriptors").fetchall():
lbry_file_id = lbry_files[stream_hash]
outpoint = name_metadata_cursor.execute("select txid, n from name_metadata "
"where sd_hash=?",
(sd_hash,)).fetchall()
if outpoint:
txid, nout = outpoint[0]
lbryfile_cursor.execute("insert into lbry_file_metadata values (?, ?, ?)",
(lbry_file_id, txid, nout))
else:
lbryfile_cursor.execute("insert into lbry_file_metadata values (?, ?, ?)",
(lbry_file_id, None, None))
lbryfile_db.commit()
lbryfile_db.close()
name_metadata_db.close()