def start_replication(self):
# TODO: We might want to read from the position where the pghoard slot is
# when restarting pghoard, instead of always reading from current position
# at the time of starting replication. The slot position is unfortunately
# not available on the replication protocol side and would have to be queried
# through a regular PG connection. Currently we workaround this by reading
# it back from pghoard's state file.
self.c.execute("IDENTIFY_SYSTEM")
identify_system = self.c.fetchone()
self.log.debug("System identified itself as: %r", identify_system)
timeline = identify_system[1]
self.fetch_timeline_history_files(timeline)
# Figure out the LSN we should try to replicate from
if self.last_flushed_lsn:
log, pos, _ = convert_integer_to_lsn(self.last_flushed_lsn)
lsn = "{:X}/{:X}".format(log, pos)
else:
lsn = get_lsn_from_start_of_wal_file(identify_system[2])
self.log.info("Starting replication from %r, timeline: %r with slot: %r",
lsn, timeline, self.replication_slot)
if self.replication_slot:
self.c.start_replication(
slot_name=self.replication_slot,
slot_type=REPLICATION_PHYSICAL,
start_lsn=lsn,
timeline=timeline)
else:
self.c.start_replication(
start_lsn=lsn,
timeline=timeline)
return timeline
def start_replication(self):
# TODO: We might want to read from the position where the pghoard slot is
# when restarting pghoard, instead of always reading from current position
# at the time of starting replication. The slot position is unfortunately
# not available on the replication protocol side and would have to be queried
# through a regular PG connection. Currently we workaround this by reading
# it back from pghoard's state file.
self.c.execute("IDENTIFY_SYSTEM")
identify_system = self.c.fetchone()
self.log.debug("System identified itself as: %r", identify_system)
timeline = identify_system[1]
self.fetch_timeline_history_files(timeline)
# Figure out the LSN we should try to replicate from
if self.last_flushed_lsn:
log, pos, _ = convert_integer_to_lsn(self.last_flushed_lsn)
lsn = "{:X}/{:X}".format(log, pos)
else:
lsn = get_lsn_from_start_of_wal_file(identify_system[2])
self.log.info(
"Starting replication from %r, timeline: %r with slot: %r", lsn,
timeline, self.replication_slot)
if self.replication_slot:
self.c.start_replication(slot_name=self.replication_slot,
slot_type=REPLICATION_PHYSICAL,
start_lsn=lsn,
timeline=timeline)
else:
self.c.start_replication(start_lsn=lsn, timeline=timeline)
return timeline
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