def delete_remote_wal_before(self, wal_segment, site):
self.log.debug("Starting WAL deletion from: %r before: %r", site, wal_segment)
storage = self.site_transfers.get(site)
valid_timeline = True
tli, log, seg = wal.name_to_tli_log_seg(wal_segment)
while True:
if valid_timeline:
# Decrement one segment if we're on a valid timeline
if seg == 0 and log == 0:
break
seg, log = wal.get_previous_wal_on_same_timeline(seg, log)
wal_path = os.path.join(self.config.get("path_prefix", ""), site, "xlog",
wal.name_for_tli_log_seg(tli, log, seg))
self.log.debug("Deleting wal_file: %r", wal_path)
try:
storage.delete_key(wal_path)
valid_timeline = True
except FileNotFoundFromStorageError:
if not valid_timeline or tli <= 1:
# if we didn't find any WALs to delete on this timeline or we're already at
# timeline 1 there's no need or possibility to try older timelines, break.
self.log.info("Could not delete wal_file: %r, returning", wal_path)
break
# let's try the same segment number on a previous timeline, but flag that timeline
# as "invalid" until we're able to delete at least one segment on it.
valid_timeline = False
tli -= 1
self.log.info("Could not delete wal_file: %r, trying the same segment on a previous "
"timeline (%s)", wal_path, wal.name_for_tli_log_seg(tli, log, seg))
except: # FIXME: don't catch all exceptions; pylint: disable=bare-except
self.log.exception("Problem deleting: %r", wal_path)
def delete_remote_wal_before(self, wal_segment, site, pg_version):
self.log.info("Starting WAL deletion from: %r before: %r, pg_version: %r",
site, wal_segment, pg_version)
storage = self.site_transfers.get(site)
valid_timeline = True
tli, log, seg = wal.name_to_tli_log_seg(wal_segment)
while True:
if valid_timeline:
# Decrement one segment if we're on a valid timeline
if seg == 0 and log == 0:
break
seg, log = wal.get_previous_wal_on_same_timeline(seg, log, pg_version)
wal_path = os.path.join(self.config["backup_sites"][site]["prefix"], "xlog",
wal.name_for_tli_log_seg(tli, log, seg))
self.log.debug("Deleting wal_file: %r", wal_path)
try:
storage.delete_key(wal_path)
valid_timeline = True
except FileNotFoundFromStorageError:
if not valid_timeline or tli <= 1:
# if we didn't find any WALs to delete on this timeline or we're already at
# timeline 1 there's no need or possibility to try older timelines, break.
self.log.info("Could not delete wal_file: %r, returning", wal_path)
break
# let's try the same segment number on a previous timeline, but flag that timeline
# as "invalid" until we're able to delete at least one segment on it.
valid_timeline = False
tli -= 1
self.log.info("Could not delete wal_file: %r, trying the same segment on a previous "
"timeline (%s)", wal_path, wal.name_for_tli_log_seg(tli, log, seg))
except Exception as ex: # FIXME: don't catch all exceptions; pylint: disable=broad-except
self.log.exception("Problem deleting: %r", wal_path)
self.metrics.unexpected_exception(ex, where="delete_remote_wal_before")
def run(self):
self._init_cursor()
if self.replication_slot:
self.create_replication_slot()
timeline = self.start_replication()
while self.running:
wal_name = None
try:
msg = self.c.read_message()
except psycopg2.DatabaseError as ex:
self.log.exception("Unexpected exception in reading walreceiver msg")
self.stats.unexpected_exception(ex, where="walreceiver_run")
continue
self.log.debug("replication_msg: %r, buffer: %r/%r",
msg, self.buffer.tell(), XLOG_SEG_SIZE)
if msg:
self.latest_activity = datetime.datetime.utcnow()
log, _, seg = convert_integer_to_lsn(msg.data_start)
wal_name = name_for_tli_log_seg(timeline, log, seg)
if not self.latest_wal:
self.latest_wal_start = msg.data_start
self.latest_wal = wal_name
self.buffer.write(msg.payload)
# TODO: Calculate end pos and transmit that?
msg.cursor.send_feedback(write_lsn=msg.data_start)
if wal_name and self.latest_wal != wal_name or self.buffer.tell() >= XLOG_SEG_SIZE:
self.switch_xlog()
for wal_start, queue in self.callbacks.items():
with suppress(Empty):
transfer_result = queue.get_nowait()
self.log.debug("Transfer result: %r", transfer_result)
self.completed_wal_segments.add(wal_start)
for completed_lsn in sorted(self.completed_wal_segments):
self.callbacks.pop(completed_lsn)
if self.callbacks:
if completed_lsn > min(self.callbacks):
pass # Do nothing since a smaller lsn is still being transferred
else: # Earlier lsn than earlist on-going transfer, just advance flush_lsn
self.c.send_feedback(flush_lsn=completed_lsn)
self.completed_wal_segments.discard(completed_lsn)
self.last_flushed_lsn = completed_lsn
self.log.debug("Sent flush_lsn feedback as: %r", self.last_flushed_lsn)
else: # No on-going transfer, just advance flush_lsn
self.c.send_feedback(flush_lsn=completed_lsn)
self.completed_wal_segments.discard(completed_lsn)
self.last_flushed_lsn = completed_lsn
self.log.debug("Sent flush_lsn feedback as: %r", self.last_flushed_lsn)
if not msg:
timeout = KEEPALIVE_INTERVAL - (datetime.datetime.now() - self.c.io_timestamp).total_seconds()
with suppress(InterruptedError):
if not any(select.select([self.c], [], [], max(0, timeout))):
self.c.send_feedback() # timing out, send keepalive
def run(self):
self._init_cursor()
if self.replication_slot:
self.create_replication_slot()
timeline = self.start_replication()
while self.running:
wal_name = None
try:
msg = self.c.read_message()
except psycopg2.DatabaseError as ex:
self.log.exception("Unexpected exception in reading walreceiver msg")
self.stats.unexpected_exception(ex, where="walreceiver_run")
continue
self.log.debug("replication_msg: %r, buffer: %r/%r",
msg, self.buffer.tell(), WAL_SEG_SIZE)
if msg:
self.latest_activity = datetime.datetime.utcnow()
log, _, seg = convert_integer_to_lsn(msg.data_start)
wal_name = name_for_tli_log_seg(timeline, log, seg)
if not self.latest_wal:
self.latest_wal_start = msg.data_start
self.latest_wal = wal_name
self.buffer.write(msg.payload)
# TODO: Calculate end pos and transmit that?
msg.cursor.send_feedback(write_lsn=msg.data_start)
if wal_name and self.latest_wal != wal_name or self.buffer.tell() >= WAL_SEG_SIZE:
self.switch_wal()
for wal_start, queue in self.callbacks.items():
with suppress(Empty):
transfer_result = queue.get_nowait()
self.log.debug("Transfer result: %r", transfer_result)
self.completed_wal_segments.add(wal_start)
for completed_lsn in sorted(self.completed_wal_segments):
self.callbacks.pop(completed_lsn)
if self.callbacks:
if completed_lsn > min(self.callbacks):
pass # Do nothing since a smaller lsn is still being transferred
else: # Earlier lsn than earlist on-going transfer, just advance flush_lsn
self.c.send_feedback(flush_lsn=completed_lsn)
self.completed_wal_segments.discard(completed_lsn)
self.last_flushed_lsn = completed_lsn
self.log.debug("Sent flush_lsn feedback as: %r", self.last_flushed_lsn)
else: # No on-going transfer, just advance flush_lsn
self.c.send_feedback(flush_lsn=completed_lsn)
self.completed_wal_segments.discard(completed_lsn)
self.last_flushed_lsn = completed_lsn
self.log.debug("Sent flush_lsn feedback as: %r", self.last_flushed_lsn)
if not msg:
timeout = KEEPALIVE_INTERVAL - (datetime.datetime.now() - self.c.io_timestamp).total_seconds()
with suppress(InterruptedError):
if not any(select.select([self.c], [], [], max(0, timeout))):
self.c.send_feedback() # timing out, send keepalive
def test_construct_wal_name():
sysinfo = {
"dbname": "",
"systemid": "6181331723016416192",
"timeline": "4",
"xlogpos": "F/190001B0",
}
assert wal.construct_wal_name(sysinfo) == wal.name_for_tli_log_seg(4, 0xF, 0x19)
assert wal.construct_wal_name(sysinfo) == "000000040000000F00000019"
def test_construct_wal_name():
sysinfo = {
"dbname": "",
"systemid": "6181331723016416192",
"timeline": "4",
"xlogpos": "F/190001B0",
}
assert wal.construct_wal_name(sysinfo) == wal.name_for_tli_log_seg(4, 0xF, 0x19)
assert wal.construct_wal_name(sysinfo) == "000000040000000F00000019"
