def test_slow_catchup_while_ordering(tdir, tconf, looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_client):
lagging_node = txnPoolNodeSet[-1]
other_lagging_node = txnPoolNodeSet[-2]
other_nodes = txnPoolNodeSet[:-1]
other_stashers = [node.nodeIbStasher for node in other_nodes]
def lagging_node_state() -> NodeLeecherService.State:
return lagging_node.ledgerManager._node_leecher._state
def check_lagging_node_is_not_syncing_audit():
assert lagging_node_state() != NodeLeecherService.State.SyncingAudit
# Prevent lagging node from ordering
with delay_rules(lagging_node.nodeIbStasher, ppDelay(), pDelay(),
cDelay()):
# Order request on all nodes except lagging one
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
# Prevent lagging node from catching up domain ledger (and finishing catchup)
with delay_rules(other_stashers, delay_domain_ledger_catchup()):
# Start catchup on lagging node
lagging_node.ledgerManager.start_catchup()
assert lagging_node_state(
) == NodeLeecherService.State.SyncingAudit
# Ensure that audit ledger is caught up by lagging node
looper.run(eventually(check_lagging_node_is_not_syncing_audit))
assert lagging_node_state() != NodeLeecherService.State.Idle
# Order one more request on all nodes except lagging one
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
# Now lagging node can catch up domain ledger which contains more transactions
# than it was when audit ledger was caught up
# Now delayed 3PC messages reach lagging node, so any transactions missed during
# catch up can be ordered, ensure that all nodes will have same data after that
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
# Ensure that even if we disable some other node pool is still functional
# (it won't be the case if old lagging node is nonfunctional)
with delay_rules(other_lagging_node.nodeIbStasher, ppDelay(), pDelay(),
cDelay()):
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
# Ensure that all nodes will eventually have same data
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
def test_commits_recvd_first(looper, txnPoolNodeSet, sdk_wallet_client,
sdk_pool_handle):
slow_node = [r.node for r in getNonPrimaryReplicas(txnPoolNodeSet, 0)][-1]
other_nodes = [n for n in txnPoolNodeSet if n != slow_node]
delay = 50
slow_node.nodeIbStasher.delay(ppDelay(delay, 0))
slow_node.nodeIbStasher.delay(pDelay(delay, 0))
sdk_send_batches_of_random_and_check(looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
num_reqs=20,
num_batches=4)
assert not slow_node.master_replica.prePrepares
assert not slow_node.master_replica.prepares
assert not slow_node.master_replica.commits
assert len(slow_node.master_replica.commitsWaitingForPrepare) > 0
slow_node.reset_delays_and_process_delayeds()
waitNodeDataEquality(looper, slow_node, *other_nodes)
assert check_if_all_equal_in_list(
[n.master_replica.ordered for n in txnPoolNodeSet])
assert slow_node.master_replica.prePrepares
assert slow_node.master_replica.prepares
assert slow_node.master_replica.commits
assert not slow_node.master_replica.commitsWaitingForPrepare
def test_backup_can_order_after_catchup(txnPoolNodeSet, looper,
sdk_pool_handle, sdk_wallet_client):
# We expect that after VC Gamma will be primary on backup
delayed_node = txnPoolNodeSet[-2]
with delay_rules_without_processing(delayed_node.nodeIbStasher,
pDelay(instId=MASTER_REPLICA_INDEX),
cDelay(instId=MASTER_REPLICA_INDEX),
ppDelay(instId=MASTER_REPLICA_INDEX)):
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, REQUEST_COUNT)
with delay_rules_without_processing(
[n.nodeIbStasher for n in txnPoolNodeSet],
old_view_pp_request_delay()):
ensure_view_change(looper, txnPoolNodeSet)
ensureElectionsDone(looper=looper, nodes=txnPoolNodeSet)
assert delayed_node.replicas._replicas[BACKUP_INST_ID].isPrimary
# Check, that backup cannot order
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, REQUEST_COUNT)
for n in txnPoolNodeSet:
assert n.replicas._replicas[BACKUP_INST_ID].last_ordered_3pc[
1] == 0
# Forcing catchup
delayed_node.start_catchup()
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
# Check, that backup can order after catchup
b_pp_seq_no_before = delayed_node.replicas._replicas[
BACKUP_INST_ID].last_ordered_3pc[1]
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, REQUEST_COUNT)
assert delayed_node.replicas._replicas[BACKUP_INST_ID].last_ordered_3pc[1] == \
b_pp_seq_no_before + REQUEST_COUNT
def test_pp_obsolescence_check_fail_for_delayed(tdir, tconf, looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client):
delay = PATCHED_ACCEPTABLE_DEVIATION_PREPREPARE_SECS + 1
lagging_node = txnPoolNodeSet[-1]
# Prevent lagging node from ordering
with delay_rules(lagging_node.nodeIbStasher, ppDelay(), pDelay(),
cDelay()):
# Order request on all nodes except lagging one
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
looper.run(asyncio.sleep(delay))
# Now delayed 3PC messages reach lagging node, so any delayed transactions
# can be processed (PrePrepare would be discarded but requested after that),
# ensure that all nodes will have same data after that
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
pp_count = get_count(lagging_node.master_replica,
lagging_node.master_replica.processPrePrepare)
assert pp_count > 0
assert get_timestamp_suspicion_count(lagging_node) == pp_count
def do_view_change_with_unaligned_prepare_certificates(
slow_nodes, nodes, looper, sdk_pool_handle, sdk_wallet_client):
"""
Perform view change with some nodes reaching lower last prepared certificate than others.
With current implementation of view change this can result with view change taking a lot of time.
"""
fast_nodes = [n for n in nodes if n not in slow_nodes]
all_stashers = [n.nodeIbStasher for n in nodes]
slow_stashers = [n.nodeIbStasher for n in slow_nodes]
# Delay some PREPAREs and all COMMITs
with delay_rules(slow_stashers, pDelay()):
with delay_rules(all_stashers, cDelay()):
# Send request
request = sdk_send_random_request(looper, sdk_pool_handle, sdk_wallet_client)
# Wait until this request is prepared on fast nodes
looper.run(eventually(check_last_prepared_certificate, fast_nodes, (0, 1)))
# Make sure its not prepared on slow nodes
looper.run(eventually(check_last_prepared_certificate, slow_nodes, None))
# Trigger view change
for n in nodes:
n.view_changer.on_master_degradation()
# Now commits are processed
# Wait until view change is complete
looper.run(eventually(check_view_change_done, nodes, 1, timeout=60))
# Finish request gracefully
sdk_get_reply(looper, request)
def test_clearing_forwarded_preprepared_request(looper, chkFreqPatched,
reqs_for_checkpoint,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_steward):
# Case when backup ordered correctly, but primary had problems.
# As a result, master will execute caughtup txns and will be removed
# from requests queues
behind_node = txnPoolNodeSet[-1]
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward,
CHK_FREQ, CHK_FREQ)
with delay_rules(behind_node.nodeIbStasher,
pDelay(delay=sys.maxsize, instId=0),
cDelay(delay=sys.maxsize, instId=0)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
assert len(behind_node.requests) == 0
assert all([
len(q) == 0 for r in behind_node.replicas.values()
for q in r.requestQueues.values()
])
assert len(behind_node.clientAuthNr._verified_reqs) == 0
assert len(behind_node.requestSender) == 0
def test_deletion_non_forwarded_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward, tconf, tdir, allPluginsPath):
behind_node = txnPoolNodeSet[-1]
[behind_node.replicas.values()[1].discard_req_key(1, key) for key in behind_node.requests]
behind_node.requests.clear()
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
behind_node.quorums.propagate = Quorum(len(txnPoolNodeSet) + 1)
with delay_rules(behind_node.nodeIbStasher,
ppDelay(delay=sys.maxsize),
pDelay(delay=sys.maxsize),
cDelay(delay=sys.maxsize)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
# We clear caughtup requests
looper.run(eventually(lambda: assertExp(len(behind_node.requests) == 0)))
assert all([len(q) == 0 for r in behind_node.replicas.values() for q in r._ordering_service.requestQueues.values()])
assert len(behind_node.clientAuthNr._verified_reqs) == 0
assert len(behind_node.requestSender) == 0
def test_revert_xfer_with_fees_before_catchup(looper, helpers,
nodeSetWithIntegratedTokenPlugin,
sdk_pool_handle, fees,
xfer_mint_tokens,
xfer_addresses):
nodes = nodeSetWithIntegratedTokenPlugin
node_stashers = [n.nodeIbStasher for n in nodes]
helpers.general.do_set_fees(fees)
[address_giver, address_receiver] = xfer_addresses
inputs = helpers.general.get_utxo_addresses([address_giver])[0]
outputs = [{
ADDRESS: address_receiver,
AMOUNT: 1000 - fees[XFER_PUBLIC_FEES_ALIAS]
}]
request = helpers.request.transfer(inputs, outputs)
with delay_rules_without_processing(node_stashers, cDelay(), pDelay()):
helpers.sdk.send_request_objects([request])
looper.runFor(waits.expectedPrePrepareTime(len(nodes)))
for n in nodes:
n.start_catchup()
for n in nodes:
looper.run(
eventually(lambda: assertExp(n.mode == Mode.participating)))
for n in nodes:
looper.run(
eventually(check_state, n, True, retryWait=0.2, timeout=15))
ensure_all_nodes_have_same_data(looper, nodes)
def test_check_cdp_pp_storages(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client):
def check_all_empty(replica):
assert not bool(replica._consensus_data.preprepared)
assert not bool(replica._consensus_data.prepared)
def check_preprepared_not_empty(replica):
assert bool(replica._consensus_data.preprepared)
def check_prepared_not_empty(replica):
assert bool(replica._consensus_data.prepared)
def operation_for_replicas(operation, node_set=txnPoolNodeSet):
for node in node_set:
operation(node.master_replica)
node_stashers = [n.nodeIbStasher for n in txnPoolNodeSet]
with delay_rules(node_stashers, pDelay()):
with delay_rules(node_stashers, ppDelay()):
sdk_send_random_request(looper, sdk_pool_handle, sdk_wallet_client)
looper.run(
eventually(operation_for_replicas, check_all_empty,
txnPoolNodeSet[1:]))
looper.run(
eventually(operation_for_replicas, check_preprepared_not_empty,
txnPoolNodeSet[0:1]))
looper.run(
eventually(operation_for_replicas, check_preprepared_not_empty,
txnPoolNodeSet))
looper.run(
eventually(operation_for_replicas, check_prepared_not_empty,
txnPoolNodeSet))
def test_freeing_forwarded_not_preprepared_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward, tconf, tdir, allPluginsPath):
behind_node = txnPoolNodeSet[-1]
behind_node.requests.clear()
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward,
CHK_FREQ, CHK_FREQ)
with delay_rules(
behind_node.nodeIbStasher,
chk_delay(delay=sys.maxsize,
instId=behind_node.replicas.values()[-1])):
with delay_rules(behind_node.nodeIbStasher, ppDelay(delay=sys.maxsize),
pDelay(delay=sys.maxsize), cDelay(delay=sys.maxsize)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(
eventually(node_caughtup, behind_node, count, retryWait=1))
looper.run(
eventually(
lambda: assertExp(len(behind_node.requests) == req_num)))
# We execute caughtup requests
looper.run(
eventually(lambda: assertExp(len(behind_node.requests) == req_num)))
assert all(r.executed for r in behind_node.requests.values()
if behind_node.seqNoDB.get(r.request.key)[1])
def test_check_cdp_pp_storages(looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_steward):
def check_all_empty(replica, reverse=False):
check_preprepared_empty(replica, reverse)
check_prepared_empty(replica, reverse)
def check_preprepared_empty(replica, reverse=False):
statement_pp = bool(replica._consensus_data.preprepared)
statement_pp ^= reverse
assert not statement_pp
def check_prepared_empty(replica, reverse=False):
statement_p = bool(replica._consensus_data.prepared)
statement_p ^= reverse
assert not statement_p
def operation_for_replicas(operation, node_set=txnPoolNodeSet, reverse=False):
for node in node_set:
operation(node.master_replica, reverse)
node_stashers = [n.nodeIbStasher for n in txnPoolNodeSet]
with delay_rules(node_stashers, cDelay()):
with delay_rules(node_stashers, pDelay()):
with delay_rules(node_stashers, ppDelay()):
sdk_add_new_nym_without_waiting(looper, sdk_pool_handle, sdk_wallet_steward)
looper.run(eventually(operation_for_replicas, check_all_empty, txnPoolNodeSet[1:]))
looper.run(eventually(operation_for_replicas, check_preprepared_empty, txnPoolNodeSet[0:1], True))
looper.run(eventually(operation_for_replicas, check_preprepared_empty, txnPoolNodeSet, True))
looper.run(eventually(operation_for_replicas, check_prepared_empty, txnPoolNodeSet, True))
looper.run(eventually(operation_for_replicas, check_all_empty, txnPoolNodeSet, True))
def test_new_primary_lagging_behind(looper, txnPoolNodeSet, sdk_wallet_client,
sdk_pool_handle, tconf):
initial_view_no = checkViewNoForNodes(txnPoolNodeSet)
next_primary_name = get_next_primary_name(txnPoolNodeSet,
initial_view_no + 1)
next_primary = [n for n in txnPoolNodeSet
if n.name == next_primary_name][0]
other_nodes = [n for n in txnPoolNodeSet if n != next_primary]
expected_primary_name = get_next_primary_name(txnPoolNodeSet,
initial_view_no + 2)
# Next primary cannot stabilize 1 checkpoint
with delay_rules(next_primary.nodeIbStasher, cDelay(), pDelay()):
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, CHK_FREQ)
ensure_view_change(looper, txnPoolNodeSet)
looper.run(
eventually(check_not_in_view_change,
txnPoolNodeSet,
timeout=2 * tconf.NEW_VIEW_TIMEOUT))
ensureElectionsDone(looper=looper,
nodes=other_nodes,
customTimeout=2 * tconf.NEW_VIEW_TIMEOUT,
instances_list=[0, 1])
assert next_primary_name != expected_primary_name
assert checkViewNoForNodes(txnPoolNodeSet) == initial_view_no + 2
# send CHK_FREQ reqs so that slow node will start catch-up
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, CHK_FREQ)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet, custom_timeout=30)
def test_deletion_non_forwarded_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward, tconf, tdir, allPluginsPath):
behind_node = txnPoolNodeSet[-1]
[behind_node.replicas.values()[1].discard_req_key(1, key) for key in behind_node.requests]
behind_node.requests.clear()
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
behind_node.quorums.propagate = Quorum(len(txnPoolNodeSet) + 1)
with delay_rules(behind_node.nodeIbStasher,
ppDelay(delay=sys.maxsize),
pDelay(delay=sys.maxsize),
cDelay(delay=sys.maxsize)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
# We clear caughtup requests
assert len(behind_node.requests) == 0
assert all([len(q) == 0 for r in behind_node.replicas.values() for q in r.requestQueues.values()])
assert len(behind_node.clientAuthNr._verified_reqs) == 0
assert len(behind_node.requestSender) == 0
def test_primary_receives_delayed_prepares(looper, txnPoolNodeSet,
sdk_wallet_client,
sdk_pool_handle):
"""
Primary gets all PREPAREs after COMMITs
"""
delay = 50
primary_node = get_master_primary_node(txnPoolNodeSet)
other_nodes = [n for n in txnPoolNodeSet if n != primary_node]
primary_node.nodeIbStasher.delay(pDelay(delay, 0))
sdk_send_random_and_check(looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
count=10)
for node in other_nodes:
assert node.master_replica.prePrepares
assert node.master_replica.prepares
assert node.master_replica.commits
assert primary_node.master_replica.sentPrePrepares
assert not primary_node.master_replica.prepares
assert primary_node.master_replica.commits
def do_view_change_with_unaligned_prepare_certificates(
slow_nodes, nodes, looper, sdk_pool_handle, sdk_wallet_client):
"""
Perform view change with some nodes reaching lower last prepared certificate than others.
With current implementation of view change this can result with view change taking a lot of time.
"""
fast_nodes = [n for n in nodes if n not in slow_nodes]
all_stashers = [n.nodeIbStasher for n in nodes]
slow_stashers = [n.nodeIbStasher for n in slow_nodes]
# Delay some PREPAREs and all COMMITs
with delay_rules(slow_stashers, pDelay()):
with delay_rules(all_stashers, cDelay()):
# Send request
request = sdk_send_random_request(looper, sdk_pool_handle, sdk_wallet_client)
# Wait until this request is prepared on fast nodes
looper.run(eventually(check_last_prepared_certificate, fast_nodes, (0, 1)))
# Make sure its not prepared on slow nodes
looper.run(eventually(check_last_prepared_certificate, slow_nodes, None))
# Trigger view change
for n in nodes:
n.view_changer.on_master_degradation()
# Now commits are processed
# Wait until view change is complete
looper.run(eventually(check_view_change_done, nodes, 1, timeout=60))
# Finish request gracefully
sdk_get_reply(looper, request)
def test_dequeue_and_validate_commits(looper, txnPoolNodeSet,
sdk_wallet_client, sdk_pool_handle):
slow_node = [r.node for r in getNonPrimaryReplicas(txnPoolNodeSet, 0)][-1]
other_nodes = [n for n in txnPoolNodeSet if n != slow_node]
delay = 50
with delay_rules(slow_node.nodeIbStasher, pDelay(delay),
msg_rep_delay(delay, [PREPARE, PREPREPARE])):
with delay_rules(slow_node.nodeIbStasher, ppDelay(delay)):
sdk_send_batches_of_random_and_check(looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
num_reqs=1,
num_batches=1)
assert not slow_node.master_replica._ordering_service.prePrepares
assert not slow_node.master_replica._ordering_service.prepares
assert not slow_node.master_replica._ordering_service.commits
assert len(slow_node.master_replica._ordering_service.
commitsWaitingForPrepare) > 0
waitNodeDataEquality(looper, slow_node, *other_nodes)
assert check_if_all_equal_in_list([
n.master_replica._ordering_service.ordered for n in txnPoolNodeSet
])
assert slow_node.master_replica._ordering_service.prePrepares
assert slow_node.master_replica._ordering_service.prepares
assert slow_node.master_replica._ordering_service.commits
assert not slow_node.master_replica._ordering_service.commitsWaitingForPrepare
assert all(slow_node.master_replica.last_ordered_3pc ==
n.master_replica.last_ordered_3pc for n in other_nodes)
def test_primary_recvs_3phase_message_outside_watermarks(perf_chk_patched, chkFreqPatched, looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_client, reqs_for_logsize):
"""
One of the primary starts getting lot of requests, more than his log size
and queues up requests since they will go beyond its watermarks. This
happens since other nodes are slow in processing its PRE-PREPARE.
Eventually this primary will send PRE-PREPARE for all requests and those
requests will complete
"""
tconf = perf_chk_patched
delay = 2
instId = 0
reqs_to_send = 2 * reqs_for_logsize + 1
logger.debug('Will send {} requests'.format(reqs_to_send))
npr = getNonPrimaryReplicas(txnPoolNodeSet, instId)
pr = getPrimaryReplica(txnPoolNodeSet, instId)
orderedCount = pr.stats.get(TPCStat.OrderSent)
for r in npr:
r.node.nodeIbStasher.delay(ppDelay(delay, instId))
r.node.nodeIbStasher.delay(pDelay(delay, instId))
tm_exec_1_batch = waits.expectedTransactionExecutionTime(len(txnPoolNodeSet))
batch_count = math.ceil(reqs_to_send / tconf.Max3PCBatchSize)
total_timeout = (tm_exec_1_batch + delay) * batch_count
def chk():
assert orderedCount + batch_count == pr.stats.get(TPCStat.OrderSent)
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_client, reqs_to_send)
looper.run(eventually(chk, retryWait=1, timeout=total_timeout))
def test_primary_recvs_3phase_message_outside_watermarks(perf_chk_patched, chkFreqPatched, looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_client, reqs_for_logsize):
"""
One of the primary starts getting lot of requests, more than his log size
and queues up requests since they will go beyond its watermarks. This
happens since other nodes are slow in processing its PRE-PREPARE.
Eventually this primary will send PRE-PREPARE for all requests and those
requests will complete
"""
tconf = perf_chk_patched
delay = 5
instId = 0
reqs_to_send = 2 * reqs_for_logsize + 1
logger.debug('Will send {} requests'.format(reqs_to_send))
npr = getNonPrimaryReplicas(txnPoolNodeSet, instId)
pr = getPrimaryReplica(txnPoolNodeSet, instId)
from plenum.server.replica import TPCStat
orderedCount = pr.stats.get(TPCStat.OrderSent)
for r in npr:
r.node.nodeIbStasher.delay(ppDelay(delay, instId))
r.node.nodeIbStasher.delay(pDelay(delay, instId))
tm_exec_1_batch = waits.expectedTransactionExecutionTime(len(txnPoolNodeSet))
batch_count = math.ceil(reqs_to_send / tconf.Max3PCBatchSize)
total_timeout = (tm_exec_1_batch + delay) * batch_count
def chk():
assert orderedCount + batch_count == pr.stats.get(TPCStat.OrderSent)
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_client, reqs_to_send)
looper.run(eventually(chk, retryWait=1, timeout=total_timeout))
def test_node_handles_forced_upgrade_on_propagate(looper, nodeSet,
sdk_pool_handle,
sdk_wallet_trustee,
validUpgradeExpForceTrue):
"""
Verifies that POOL_UPGRADE force=true request is handled immediately when
the node receives it in a PROPAGATE from any other node
"""
slow_node = getNonPrimaryReplicas(nodeSet, instId=0)[-1].node
# Stash all except PROPAGATEs from Gamma
slow_node.clientIbStasher.delay(req_delay())
slow_node.nodeIbStasher.delay(ppgDelay(sender_filter='Alpha'))
slow_node.nodeIbStasher.delay(ppgDelay(sender_filter='Beta'))
slow_node.nodeIbStasher.delay(ppDelay())
slow_node.nodeIbStasher.delay(pDelay())
slow_node.nodeIbStasher.delay(cDelay())
sdk_send_upgrade(looper, sdk_pool_handle, sdk_wallet_trustee,
validUpgradeExpForceTrue)
looper.run(
eventually(checkUpgradeScheduled, [slow_node],
validUpgradeExpForceTrue[VERSION],
retryWait=1,
timeout=waits.expectedUpgradeScheduled()))
def test_freeing_forwarded_preprepared_request(looper, chkFreqPatched,
reqs_for_checkpoint,
txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward):
# Case, when both backup and primary had problems
behind_node = txnPoolNodeSet[-1]
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward,
CHK_FREQ, CHK_FREQ)
with delay_rules(
behind_node.nodeIbStasher,
pDelay(delay=sys.maxsize),
cDelay(delay=sys.maxsize),
):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
looper.run(
eventually(lambda: assertExp(len(behind_node.requests) == req_num)))
assert all(r.executed for r in behind_node.requests.values()
if behind_node.seqNoDB.get(r.request.key)[1])
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward,
CHK_FREQ, CHK_FREQ)
# Master and backup replicas do not stash new requests and succesfully order them
assert len(behind_node.requests) == req_num
def test_revert_nym_with_fees_before_catchup(looper, helpers,
nodeSetWithIntegratedTokenPlugin,
fees_set, fees, xfer_mint_tokens,
xfer_addresses):
nodes = nodeSetWithIntegratedTokenPlugin
current_amount = get_amount_from_token_txn(xfer_mint_tokens)
seq_no = get_seq_no(xfer_mint_tokens)
lagging_node = nodes[-1]
current_amount, seq_no, _ = send_and_check_nym_with_fees(
helpers, fees_set, seq_no, looper, xfer_addresses, current_amount)
with delay_rules_without_processing(lagging_node.nodeIbStasher, cDelay(),
pDelay()):
current_amount, seq_no, _ = send_and_check_nym_with_fees(
helpers, fees_set, seq_no, looper, xfer_addresses, current_amount)
looper.runFor(waits.expectedPrePrepareTime(len(nodes)))
lagging_node.start_catchup()
for n in nodes:
looper.run(
eventually(lambda: assertExp(n.mode == Mode.participating)))
for n in nodes:
looper.run(
eventually(check_state, n, True, retryWait=0.2, timeout=15))
ensure_all_nodes_have_same_data(looper, nodes)
current_amount, seq_no, _ = send_and_check_nym_with_fees(
helpers, fees_set, seq_no, looper, xfer_addresses, current_amount)
ensure_all_nodes_have_same_data(looper, nodes)
def test_no_propagate_request_on_different_prepares_on_backup_before_vc(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_client):
'''
1. Send random request
2. Make 3 node on backup instance slow in getting prepares
3. Send random request
4. do view change
5. reset delays
=> we expect that all nodes and all instances have the same last ordered
'''
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
slow_instance = 1
slow_nodes = txnPoolNodeSet[1:3]
fast_nodes = [n for n in txnPoolNodeSet if n not in slow_nodes]
nodes_stashers = [n.nodeIbStasher for n in slow_nodes]
old_last_ordered = txnPoolNodeSet[0].master_replica.last_ordered_3pc
with delay_rules(nodes_stashers, pDelay(instId=slow_instance)):
with delay_rules(nodes_stashers, ppDelay(instId=slow_instance)):
# send one request
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
old_view_no = txnPoolNodeSet[0].viewNo
looper.run(
eventually(is_prepared,
fast_nodes,
2,
slow_instance))
# trigger view change on all nodes
ensure_view_change(looper, txnPoolNodeSet)
# wait for view change done on all nodes
ensureElectionsDone(looper, txnPoolNodeSet)
primary = getPrimaryReplica(txnPoolNodeSet, slow_instance).node
non_primaries = [n for n in txnPoolNodeSet if n is not primary]
check_last_ordered(non_primaries,
slow_instance,
(old_view_no, old_last_ordered[1] + 1))
# Backup primary replica must not advance last_ordered_3pc
# up to the master's value
check_last_ordered([primary],
slow_instance,
(old_view_no, old_last_ordered[1]))
check_last_ordered(txnPoolNodeSet,
txnPoolNodeSet[0].master_replica.instId,
(old_last_ordered[0], old_last_ordered[1] + 1))
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
looper.run(
eventually(check_last_ordered,
txnPoolNodeSet,
slow_instance,
(txnPoolNodeSet[0].viewNo, 1)))
assert all(0 == node.spylog.count(node.request_propagates)
for node in txnPoolNodeSet)
def test_freeing_forwarded_preprepared_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward):
# Case, when both backup and primary had problems
behind_node = txnPoolNodeSet[-1]
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
with delay_rules(behind_node.nodeIbStasher,
pDelay(delay=sys.maxsize),
cDelay(delay=sys.maxsize), ):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
assert len(behind_node.requests) == req_num
assert all(r.executed for r in behind_node.requests.values() if behind_node.seqNoDB.get(r.request.key)[1])
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
# Master and backup replicas do not stash new requests and succesfully order them
assert len(behind_node.requests) == req_num
def test_commits_recvd_first(looper, txnPoolNodeSet,
sdk_wallet_client, sdk_pool_handle):
slow_node = [r.node for r in getNonPrimaryReplicas(txnPoolNodeSet, 0)][-1]
other_nodes = [n for n in txnPoolNodeSet if n != slow_node]
delay = 50
slow_node.nodeIbStasher.delay(ppDelay(delay, 0))
slow_node.nodeIbStasher.delay(pDelay(delay, 0))
sdk_send_batches_of_random_and_check(looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
num_reqs=20,
num_batches=4)
assert not slow_node.master_replica.prePrepares
assert not slow_node.master_replica.prepares
assert not slow_node.master_replica.commits
assert len(slow_node.master_replica.commitsWaitingForPrepare) > 0
slow_node.reset_delays_and_process_delayeds()
waitNodeDataEquality(looper, slow_node, *other_nodes)
assert check_if_all_equal_in_list([n.master_replica.ordered
for n in txnPoolNodeSet])
assert slow_node.master_replica.prePrepares
assert slow_node.master_replica.prepares
assert slow_node.master_replica.commits
assert not slow_node.master_replica.commitsWaitingForPrepare
def setup(request, looper, txnPoolNodeSet, client1, wallet1, client1Connected):
slow_node = getNonPrimaryReplicas(txnPoolNodeSet, 0)[1].node
fast_nodes = [n for n in txnPoolNodeSet if n != slow_node]
slow_node.nodeIbStasher.delay(pDelay(100, 0))
slow_node.nodeIbStasher.delay(cDelay(100, 0))
if request.param == 'all':
slow_node.nodeIbStasher.delay(ppDelay(100, 0))
return slow_node, fast_nodes
def test_view_change_after_max_catchup_rounds(txnPoolNodeSet, looper, sdk_pool_handle, sdk_wallet_client):
"""
The node should do only a fixed rounds of catchup. For this delay Prepares
and Commits for 2 non-primary nodes by a large amount which is equivalent
to loss of Prepares and Commits. Make sure 2 nodes have a different last
prepared certificate from other two. Then do a view change, make sure view
change completes and the pool does not process the request that were
prepared by only a subset of the nodes
"""
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 2 * 3, 3)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
ledger_summary = txnPoolNodeSet[0].ledger_summary
slow_nodes = [r.node for r in getNonPrimaryReplicas(
txnPoolNodeSet, 0)[-2:]]
fast_nodes = [n for n in txnPoolNodeSet if n not in slow_nodes]
# Make node slow to process Prepares and Commits
for node in slow_nodes:
node.nodeIbStasher.delay(pDelay(120, 0))
node.nodeIbStasher.delay(cDelay(120, 0))
sdk_send_random_requests(looper, sdk_pool_handle, sdk_wallet_client, 5)
looper.runFor(3)
ensure_view_change(looper, nodes=txnPoolNodeSet)
def last_prepared(nodes):
lst = [n.master_replica.last_prepared_certificate_in_view()
for n in nodes]
# All nodes have same last prepared
assert check_if_all_equal_in_list(lst)
return lst[0]
last_prepared_slow = last_prepared(slow_nodes)
last_prepared_fast = last_prepared(fast_nodes)
# Check `slow_nodes` and `fast_nodes` set different last_prepared
assert last_prepared_fast != last_prepared_slow
# View change complete
ensureElectionsDone(looper, txnPoolNodeSet)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
# The requests which were prepared by only a subset of the nodes were
# not ordered
assert txnPoolNodeSet[0].ledger_summary == ledger_summary
for node in slow_nodes:
node.nodeIbStasher.reset_delays_and_process_delayeds()
# Make sure pool is functional
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 10, 2)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
last_prepared(txnPoolNodeSet)
def test_revert_auth_rule_changing(looper, txnPoolNodeSet, sdk_wallet_trustee,
sdk_wallet_steward, sdk_pool_handle):
node_stashers = [n.nodeIbStasher for n in txnPoolNodeSet]
wh, _ = sdk_wallet_trustee
new_steward_did, new_steward_verkey = create_verkey_did(looper, wh)
new_steward_did2, new_steward_verkey2 = create_verkey_did(looper, wh)
"""We try to change rule for adding new steward. For this case we """
changed_constraint = AuthConstraint(role=STEWARD, sig_count=1)
action = AuthActionAdd(txn_type=NYM, field=ROLE, value=STEWARD)
with delay_rules_without_processing(node_stashers, pDelay(), cDelay()):
sdk_send_and_check_auth_rule_request(
looper,
sdk_wallet_trustee,
sdk_pool_handle,
auth_action=ADD_PREFIX,
auth_type=action.txn_type,
field=action.field,
new_value=action.value,
old_value=None,
constraint=changed_constraint.as_dict,
no_wait=True)
looper.runFor(waits.expectedPrePrepareTime(len(txnPoolNodeSet)))
"""
Try to add new steward by already existed trustee.
Validation should raise exception because we change uncommitted state
by adding new rule, that "Only steward can add new steward"
"""
with pytest.raises(RequestRejectedException,
match="TRUSTEE can not do this action"):
sdk_add_new_nym(looper,
sdk_pool_handle,
sdk_wallet_trustee,
'newSteward1',
STEWARD_STRING,
dest=new_steward_did,
verkey=new_steward_verkey)
looper.runFor(waits.expectedPrePrepareTime(len(txnPoolNodeSet)))
"""
Catchup should revert config_state and discard rule changing
"""
for n in txnPoolNodeSet:
n.start_catchup()
for n in txnPoolNodeSet:
looper.run(
eventually(lambda: assertExp(n.mode == Mode.participating)))
"""
Try to create new steward by steward
We can not do this, because AUTH_RULE txn was reverted
"""
with pytest.raises(RequestRejectedException,
match="STEWARD can not do this action"):
sdk_add_new_nym(looper,
sdk_pool_handle,
sdk_wallet_steward,
'newSteward2',
STEWARD_STRING,
dest=new_steward_did2,
verkey=new_steward_verkey2)
def test_view_change_after_max_catchup_rounds(txnPoolNodeSet, looper, wallet1,
client1, client1Connected):
"""
The node should do only a fixed rounds of catchup. For this delay Prepares
and Commits for 2 non-primary nodes by a large amount which is equivalent
to loss of Prepares and Commits. Make sure 2 nodes have a different last
prepared certificate from other two. Then do a view change, make sure view
change completes and the pool does not process the request that were
prepared by only a subset of the nodes
"""
send_reqs_batches_and_get_suff_replies(looper, wallet1, client1, 2 * 3, 3)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
ledger_summary = txnPoolNodeSet[0].elector.ledger_summary
slow_nodes = [
r.node for r in getNonPrimaryReplicas(txnPoolNodeSet, 0)[-2:]
]
fast_nodes = [n for n in txnPoolNodeSet if n not in slow_nodes]
# Make node slow to process Prepares and Commits
for node in slow_nodes:
node.nodeIbStasher.delay(pDelay(120, 0))
node.nodeIbStasher.delay(cDelay(120, 0))
sendRandomRequests(wallet1, client1, 5)
looper.runFor(3)
ensure_view_change(looper, nodes=txnPoolNodeSet)
def last_prepared(nodes):
lst = [
n.master_replica.last_prepared_certificate_in_view() for n in nodes
]
# All nodes have same last prepared
assert check_if_all_equal_in_list(lst)
return lst[0]
last_prepared_slow = last_prepared(slow_nodes)
last_prepared_fast = last_prepared(fast_nodes)
# Check `slow_nodes` and `fast_nodes` set different last_prepared
assert last_prepared_fast != last_prepared_slow
# View change complete
ensureElectionsDone(looper, txnPoolNodeSet)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
# The requests which were prepared by only a subset of the nodes were
# not ordered
assert txnPoolNodeSet[0].elector.ledger_summary == ledger_summary
for node in slow_nodes:
node.nodeIbStasher.reset_delays_and_process_delayeds()
# Make sure pool is functional
ensure_pool_functional(looper, txnPoolNodeSet, wallet1, client1)
ensure_all_nodes_have_same_data(looper, txnPoolNodeSet)
last_prepared(txnPoolNodeSet)
def test_no_propagate_request_on_different_prepares_on_backup_before_vc(looper, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_client):
''' Send random request and do view change then fast_nodes (2,3 - with
primary backup replica) will have prepare or send preprepare on backup
replicas and slow_nodes are have not and transaction will ordered on all
master replicas. Check last ordered after view change and after another
one request.'''
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
slow_instance = 1
slow_nodes = txnPoolNodeSet[1:3]
fast_nodes = [n for n in txnPoolNodeSet if n not in slow_nodes]
nodes_stashers = [n.nodeIbStasher for n in slow_nodes]
old_last_ordered = txnPoolNodeSet[0].master_replica.last_ordered_3pc
with delay_rules(nodes_stashers, pDelay(instId=slow_instance)):
with delay_rules(nodes_stashers, ppDelay(instId=slow_instance)):
# send one request
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
old_view_no = txnPoolNodeSet[0].viewNo
looper.run(
eventually(is_prepared,
fast_nodes,
2,
slow_instance))
# trigger view change on all nodes
ensure_view_change(looper, txnPoolNodeSet)
# wait for view change done on all nodes
ensureElectionsDone(looper, txnPoolNodeSet)
primary = getPrimaryReplica(txnPoolNodeSet, slow_instance).node
non_primaries = [n for n in txnPoolNodeSet if n is not primary]
check_last_ordered(non_primaries,
slow_instance,
(old_view_no, old_last_ordered[1] + 1))
# Backup primary replica must not advance last_ordered_3pc
# up to the master's value
check_last_ordered([primary],
slow_instance,
(old_view_no, old_last_ordered[1]))
check_last_ordered(txnPoolNodeSet,
txnPoolNodeSet[0].master_replica.instId,
(old_last_ordered[0], old_last_ordered[1] + 1))
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 1)
looper.run(
eventually(check_last_ordered,
txnPoolNodeSet,
slow_instance,
(txnPoolNodeSet[0].viewNo, 1)))
assert all(0 == node.spylog.count(node.request_propagates)
for node in txnPoolNodeSet)
def test_re_order_pre_prepares(looper, txnPoolNodeSet, sdk_wallet_client,
sdk_pool_handle):
# 0. use new 3PC validator
for n in txnPoolNodeSet:
ordering_service = n.master_replica._ordering_service
ordering_service._validator = OrderingServiceMsgValidator(
ordering_service._data)
# 1. drop Prepares and Commits on 4thNode
# Order a couple of requests on Nodes 1-3
lagging_node = txnPoolNodeSet[-1]
other_nodes = txnPoolNodeSet[:-1]
with delay_rules_without_processing(lagging_node.nodeIbStasher, cDelay(),
pDelay()):
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_client, 3)
assert all(n.master_last_ordered_3PC == (0, 3) for n in other_nodes)
# 2. simulate view change start so that
# all PrePrepares/Prepares/Commits are cleared
# and uncommitted txns are reverted
for n in txnPoolNodeSet:
n.replicas.send_to_internal_bus(ViewChangeStarted(view_no=1))
master_ordering_service = n.master_replica._ordering_service
assert not master_ordering_service.prePrepares
assert not master_ordering_service.prepares
assert not master_ordering_service.commits
assert master_ordering_service.old_view_preprepares
ledger = n.db_manager.ledgers[DOMAIN_LEDGER_ID]
state = n.db_manager.states[DOMAIN_LEDGER_ID]
assert len(ledger.uncommittedTxns) == 0
assert ledger.uncommitted_root_hash == ledger.tree.root_hash
assert state.committedHead == state.head
# 3. Simulate View Change finish to re-order the same PrePrepare
assert lagging_node.master_last_ordered_3PC == (0, 0)
new_master = txnPoolNodeSet[1]
batches = sorted([
preprepare_to_batch_id(pp) for _, pp in new_master.master_replica.
_ordering_service.old_view_preprepares.items()
])
new_view_msg = NewViewCheckpointsApplied(view_no=0,
view_changes=[],
checkpoint=None,
batches=batches)
for n in txnPoolNodeSet:
n.master_replica._consensus_data.prev_view_prepare_cert = batches[
-1].pp_seq_no
n.master_replica._ordering_service._bus.send(new_view_msg)
# 4. Make sure that the nodes 1-3 (that already ordered the requests) sent Prepares and Commits so that
# the request was eventually ordered on Node4 as well
waitNodeDataEquality(looper, lagging_node, *other_nodes)
assert lagging_node.master_last_ordered_3PC == (0, 4)
sdk_ensure_pool_functional(looper, txnPoolNodeSet, sdk_wallet_client,
sdk_pool_handle)
def testNonPrimaryRecvs3PhaseMessageOutsideWatermarks(chkFreqPatched, looper,
txnPoolNodeSet, client1,
wallet1,
client1Connected,
reqs_for_logsize):
"""
A node is slow in processing PRE-PREPAREs and PREPAREs such that lot of
requests happen and the slow node has started getting 3 phase messages
outside of it watermarks. Check that it queues up requests outside watermarks and once it
has received stable checkpoint it processes more requests. It sends other
nodes 3 phase messages older than their stable checkpoint so they should
discard them.
"""
delay = 15
instId = 1
reqsToSend = reqs_for_logsize + 2
npr = getNonPrimaryReplicas(txnPoolNodeSet, instId)
slowReplica = npr[0]
slowNode = slowReplica.node
slowNode.nodeIbStasher.delay(ppDelay(delay, instId))
slowNode.nodeIbStasher.delay(pDelay(delay, instId))
def discardCounts(replicas, pat):
counts = {}
for r in replicas:
counts[r.name] = countDiscarded(r, pat)
return counts
oldStashCount = slowReplica.spylog.count(
TestReplica.stashOutsideWatermarks.__name__)
oldDiscardCounts = discardCounts(
[n.replicas[instId] for n in txnPoolNodeSet if n != slowNode],
'achieved stable checkpoint')
sendReqsToNodesAndVerifySuffReplies(looper, wallet1, client1, reqsToSend,
1)
timeout = waits.expectedPoolGetReadyTimeout(len(txnPoolNodeSet))
looper.run(
eventually(checkNodeDataForEquality,
slowNode,
*[_ for _ in txnPoolNodeSet if _ != slowNode],
retryWait=1,
timeout=timeout))
newStashCount = slowReplica.spylog.count(
TestReplica.stashOutsideWatermarks.__name__)
assert newStashCount > oldStashCount
def chk():
counts = discardCounts(
[n.replicas[instId] for n in txnPoolNodeSet if n != slowNode],
'achieved stable checkpoint')
for nm, count in counts.items():
assert count > oldDiscardCounts[nm]
timeout = waits.expectedNodeToNodeMessageDeliveryTime() * \
len(txnPoolNodeSet) + delay
looper.run(eventually(chk, retryWait=1, timeout=timeout))
def test_node_request_preprepare(looper, txnPoolNodeSet, client1,
wallet1, client1Connected, teardown):
"""
Node requests PRE-PREPARE only once after getting PREPAREs.
"""
slow_node, other_nodes, primary_node, \
other_primary_nodes = split_nodes(txnPoolNodeSet)
# Drop PrePrepares and Prepares
slow_node.nodeIbStasher.delay(ppDelay(300, 0))
slow_node.nodeIbStasher.delay(pDelay(300, 0))
send_reqs_batches_and_get_suff_replies(looper, wallet1, client1, 10, 5)
slow_node.nodeIbStasher.drop_delayeds()
slow_node.nodeIbStasher.resetDelays()
old_count_req = count_requested_preprepare_req(slow_node)
old_count_resp = count_requested_preprepare_resp(slow_node)
def chk(increase=True):
# Method is called
assert count_requested_preprepare_req(slow_node) > old_count_req
# Requesting Preprepare
assert count_requested_preprepare_resp(
slow_node) - old_count_resp == (1 if increase else 0)
for pp in primary_node.master_replica.sentPrePrepares.values():
for rep in [n.master_replica for n in other_primary_nodes]:
prepare = Prepare(rep.instId,
pp.viewNo,
pp.ppSeqNo,
pp.ppTime,
pp.digest,
pp.stateRootHash,
pp.txnRootHash
)
rep.send(prepare)
looper.run(eventually(chk, True, retryWait=1))
old_count_resp = count_requested_preprepare_resp(slow_node)
prepare = Prepare(rep.instId,
pp.viewNo,
pp.ppSeqNo,
pp.ppTime,
pp.digest,
pp.stateRootHash,
pp.txnRootHash
)
rep.send(prepare)
looper.run(eventually(chk, False, retryWait=1))
old_count_req = count_requested_preprepare_req(slow_node)
old_count_resp = count_requested_preprepare_resp(slow_node)
def test_revert_for_all_after_view_change(looper, helpers,
nodeSetWithIntegratedTokenPlugin,
sdk_pool_handle, fees_set,
mint_tokens, addresses, fees):
node_set = nodeSetWithIntegratedTokenPlugin
current_amount = get_amount_from_token_txn(mint_tokens)
seq_no = get_seq_no(mint_tokens)
reverted_node = nodeSetWithIntegratedTokenPlugin[-1]
current_amount, seq_no, _ = send_and_check_nym_with_fees(
helpers, fees_set, seq_no, looper, addresses, current_amount)
current_amount, seq_no, _ = send_and_check_transfer(
helpers, addresses, fees, looper, current_amount, seq_no)
ensure_all_nodes_have_same_data(looper, node_set)
with delay_rules([n.nodeIbStasher for n in node_set], cDelay(), pDelay()):
len_batches_before = len(reverted_node.master_replica.batches)
current_amount, seq_no, resp1 = send_and_check_transfer(
helpers,
addresses,
fees,
looper,
current_amount,
seq_no,
check_reply=False)
current_amount, seq_no, resp2 = send_and_check_nym_with_fees(
helpers,
fees_set,
seq_no,
looper,
addresses,
current_amount,
check_reply=False)
looper.runFor(
waits.expectedPrePrepareTime(
len(nodeSetWithIntegratedTokenPlugin)))
len_batches_after = len(reverted_node.master_replica.batches)
"""
Checks, that we have a 2 new batches
"""
assert len_batches_after - len_batches_before == 2
for n in node_set:
n.view_changer.on_master_degradation()
ensure_view_change_complete(looper, nodeSetWithIntegratedTokenPlugin)
looper.run(
eventually(
lambda: assertExp(reverted_node.mode == Mode.participating)))
ensure_all_nodes_have_same_data(looper, node_set)
sdk_get_and_check_replies(looper, resp1)
sdk_get_and_check_replies(looper, resp2)
send_and_check_nym_with_fees(helpers, fees_set, seq_no, looper, addresses,
current_amount)
ensure_all_nodes_have_same_data(looper, node_set)
def make_master_replica_lag(node):
# class AbysmalBox(list):
# def append(self, object) -> None:
# pass
#
# node.replicas._master_replica.inBox = AbysmalBox()
node.nodeIbStasher.delay(ppDelay(1200, 0))
node.nodeIbStasher.delay(pDelay(1200, 0))
node.nodeIbStasher.delay(cDelay(1200, 0))
def test_backups_dont_order_while_reordering(txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
looper):
"""
This test needs to show that for now we stop ordering on backups
until master in reordering state after view_change
Steps:
1. Delay ordering on master replica for collecting requests to reorder after VC
2. Make sure that master didn't order
3. Delay old_view_pp_request and force VC
4. Ensure that all backup replica on all nodes cannot order
because primary waiting for reordering on master
"""
def check_pp_count(node, expected_count, inst_id=0):
assert node.replicas._replicas[inst_id].last_ordered_3pc[1] == expected_count, \
"master last ordered: {}, backup_last_ordered: {}".format(node.master_replica._ordering_service.batches,
node.replicas._replicas[
1]._ordering_service.batches)
# We expect that after VC Gamma will be primary on backup
delayed_node = txnPoolNodeSet[-2]
fast_nodes = [n for n in txnPoolNodeSet if n != delayed_node]
master_pp_seq_no_before = delayed_node.master_replica.last_ordered_3pc[1]
with delay_rules_without_processing(delayed_node.nodeIbStasher,
pDelay(instId=MASTER_REPLICA_INDEX),
cDelay(instId=MASTER_REPLICA_INDEX),
msg_req_delay(),
msg_rep_delay(),
ppDelay(instId=MASTER_REPLICA_INDEX)):
sdk_send_random_requests(looper, sdk_pool_handle, sdk_wallet_client, REQS_FOR_REORDERING)
looper.run(eventually(check_pp_count, delayed_node, REQS_FOR_REORDERING, BACKUP_INST_ID))
assert delayed_node.master_replica.last_ordered_3pc[1] == master_pp_seq_no_before
with delay_rules([n.nodeIbStasher for n in txnPoolNodeSet], old_view_pp_request_delay()):
ensure_view_change(looper, txnPoolNodeSet)
# check that view change is finished on all nodes
looper.run(eventually(check_not_in_view_change, txnPoolNodeSet))
# check that delayed node is selected on all fast nodes but not on delayed node
def check_backup_primaries():
assert delayed_node.replicas[BACKUP_INST_ID]._consensus_data.primary_name is None
assert delayed_node.master_replica.last_ordered_3pc[1] == master_pp_seq_no_before
assert all(
n.replicas[BACKUP_INST_ID]._consensus_data.primary_name == generateName(delayed_node.name,
instId=BACKUP_INST_ID)
for n in fast_nodes
)
looper.run(eventually(check_backup_primaries))
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_client, REQS_FOR_REORDERING)
for node in txnPoolNodeSet:
assert node.replicas._replicas[BACKUP_INST_ID].last_ordered_3pc[1] == 0
def setup(request, looper, txnPoolNodeSet):
slow_node = getNonPrimaryReplicas(txnPoolNodeSet, 0)[1].node
fast_nodes = [n for n in txnPoolNodeSet if n != slow_node]
# Delay catchup reply so that the test gets time to make the check,
# this delay is reset after the check
slow_node.nodeIbStasher.delay(cr_delay(100))
slow_node.nodeIbStasher.delay(pDelay(100, 0))
slow_node.nodeIbStasher.delay(cDelay(100, 0))
if request.param == 'all':
slow_node.nodeIbStasher.delay(ppDelay(100, 0))
return slow_node, fast_nodes
def setup(request, looper, txnPoolNodeSet):
slow_node = getNonPrimaryReplicas(txnPoolNodeSet, 0)[1].node
fast_nodes = [n for n in txnPoolNodeSet if n != slow_node]
# Delay catchup reply so that the test gets time to make the check,
# this delay is reset after the check
slow_node.nodeIbStasher.delay(cr_delay(100))
slow_node.nodeIbStasher.delay(pDelay(100, 0))
slow_node.nodeIbStasher.delay(cDelay(100, 0))
if request.param == 'all':
slow_node.nodeIbStasher.delay(ppDelay(100, 0))
return slow_node, fast_nodes
def testOrderingWhenPrePrepareNotReceived(looper, txnPoolNodeSet,
sdk_wallet_client, sdk_pool_handle):
"""
Send commits but delay pre-prepare and prepares such that enough
commits are received, now the request should not be ordered until
pre-prepare is received and ordering should just happen once,
"""
delay = 10
non_prim_reps = getNonPrimaryReplicas(txnPoolNodeSet, 0)
slow_rep = non_prim_reps[0]
slow_node = slow_rep.node
slow_node.nodeIbStasher.delay(ppDelay(delay, 0))
slow_node.nodeIbStasher.delay(pDelay(delay, 0))
stash_pp = []
stash_p = []
orig_pp_method = slow_rep._ordering_service.process_preprepare
orig_p_method = slow_rep._ordering_service.process_prepare
def patched_pp(self, msg, sender):
stash_pp.append((msg, sender))
def patched_p(self, msg, sender):
stash_p.append((msg, sender))
slow_rep._ordering_service.process_preprepare = \
types.MethodType(patched_pp, slow_rep)
slow_rep._ordering_service.process_prepare = \
types.MethodType(patched_p, slow_rep)
def chk1():
assert len(slow_rep._ordering_service.commitsWaitingForPrepare) > 0
sdk_send_random_request(looper, sdk_pool_handle, sdk_wallet_client)
timeout = waits.expectedPrePrepareTime(len(txnPoolNodeSet)) + delay
looper.run(eventually(chk1, retryWait=1, timeout=timeout))
for m, s in stash_pp:
orig_pp_method(m, s)
for m, s in stash_p:
orig_p_method(m, s)
def chk2():
assert len(slow_rep._ordering_service.commitsWaitingForPrepare) == 0
assert slow_rep._ordering_service.spylog.count(
slow_rep._ordering_service._do_order.__name__) == 1
timeout = waits.expectedOrderingTime(len(non_prim_reps) + 1) + 2 * delay
looper.run(eventually(chk2, retryWait=1, timeout=timeout))
def scenario_txns_during_view_change(looper,
helpers,
nodes,
io_addresses,
send_txns,
send_txns_invalid=None):
lagging_node = nodes[-1]
rest_nodes = nodes[:-1]
def send_txns_invalid_default():
# TODO non-public API is used
addr = helpers.wallet.address_map[io_addresses()[0][0]]
seq_no = list(addr.outputs[0])[0]
assert addr.outputs[0][seq_no] > 0
# de-sync client-server utxos states
addr.outputs[0][seq_no] += 1
with pytest.raises(RequestRejectedException,
match='Insufficient funds'):
send_txns()
# sync back client-server utxos states
addr.outputs[0][seq_no] -= 1
# Send transactions
send_txns()
ensure_all_nodes_have_same_data(looper, nodes)
# Lag one node (delay Prepare and Commit messages for lagging_node)
with delay_rules(lagging_node.nodeIbStasher, pDelay(), cDelay()):
# Send more transactions
send_txns()
ensure_all_nodes_have_same_data(looper, rest_nodes)
# Send invalid transactions
(send_txns_invalid or send_txns_invalid_default)()
ensure_all_nodes_have_same_data(looper, rest_nodes)
# Initiate view change
# Wait until view change is finished and check that needed transactions are written.
ensure_view_change(looper, nodes)
looper.run(eventually(check_not_in_view_change, nodes))
ensureElectionsDone(looper, nodes)
# Reset delays
# Make sure that all nodes have equal state
# (expecting that lagging_node caught up missed ones)
ensure_all_nodes_have_same_data(looper, nodes)
# make sure the poll is functional
send_txns()
ensure_all_nodes_have_same_data(looper, nodes)
def testOrderingWhenPrePrepareNotReceived(looper, txnPoolNodeSet,
sdk_wallet_client, sdk_pool_handle):
"""
Send commits but delay pre-prepare and prepares such that enough
commits are received, now the request should not be ordered until
pre-prepare is received and ordering should just happen once,
"""
delay = 10
non_prim_reps = getNonPrimaryReplicas(txnPoolNodeSet, 0)
slow_rep = non_prim_reps[0]
slow_node = slow_rep.node
slow_node.nodeIbStasher.delay(ppDelay(delay, 0))
slow_node.nodeIbStasher.delay(pDelay(delay, 0))
stash_pp = []
stash_p = []
orig_pp_method = slow_rep.processPrePrepare
orig_p_method = slow_rep.processPrepare
def patched_pp(self, msg, sender):
stash_pp.append((msg, sender))
def patched_p(self, msg, sender):
stash_p.append((msg, sender))
slow_rep.processPrePrepare = \
types.MethodType(patched_pp, slow_rep)
slow_rep.processPrepare = \
types.MethodType(patched_p, slow_rep)
def chk1():
assert len(slow_rep.commitsWaitingForPrepare) > 0
sdk_send_random_request(looper, sdk_pool_handle, sdk_wallet_client)
timeout = waits.expectedPrePrepareTime(len(txnPoolNodeSet)) + delay
looper.run(eventually(chk1, retryWait=1, timeout=timeout))
for m, s in stash_pp:
orig_pp_method(m, s)
for m, s in stash_p:
orig_p_method(m, s)
def chk2():
assert len(slow_rep.commitsWaitingForPrepare) == 0
assert slow_rep.spylog.count(slow_rep.doOrder.__name__) == 1
timeout = waits.expectedOrderingTime(len(non_prim_reps) + 1) + 2 * delay
looper.run(eventually(chk2, retryWait=1, timeout=timeout))
def test_freeing_forwarded_not_preprepared_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward, tconf, tdir, allPluginsPath):
behind_node = txnPoolNodeSet[-1]
behind_node.requests.clear()
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
with delay_rules(behind_node.nodeIbStasher,
ppDelay(delay=sys.maxsize),
pDelay(delay=sys.maxsize),
cDelay(delay=sys.maxsize)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
# We execute caughtup requests
assert len(behind_node.requests) == req_num
assert all(r.executed for r in behind_node.requests.values() if behind_node.seqNoDB.get(r.request.key)[1])
def test_clearing_forwarded_preprepared_request(
looper, chkFreqPatched, reqs_for_checkpoint, txnPoolNodeSet,
sdk_pool_handle, sdk_wallet_steward):
# Case when backup ordered correctly, but primary had problems.
# As a result, master will execute caughtup txns and will be removed
# from requests queues
behind_node = txnPoolNodeSet[-1]
sdk_send_batches_of_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, CHK_FREQ, CHK_FREQ)
with delay_rules(behind_node.nodeIbStasher,
pDelay(delay=sys.maxsize, instId=0),
cDelay(delay=sys.maxsize, instId=0)):
count = behind_node.spylog.count(behind_node.allLedgersCaughtUp)
sdk_send_batches_of_random(looper, txnPoolNodeSet, sdk_pool_handle,
sdk_wallet_steward, req_num, req_num)
looper.run(eventually(node_caughtup, behind_node, count, retryWait=1))
assert len(behind_node.requests) == 0
assert all([len(q) == 0 for r in behind_node.replicas.values() for q in r.requestQueues.values()])
assert len(behind_node.clientAuthNr._verified_reqs) == 0
assert len(behind_node.requestSender) == 0
def make_master_replica_lag(node):
node.nodeIbStasher.delay(ppDelay(1200, 0))
node.nodeIbStasher.delay(pDelay(1200, 0))
node.nodeIbStasher.delay(cDelay(1200, 0))
def test_node_request_preprepare(looper, txnPoolNodeSet,
sdk_wallet_client, sdk_pool_handle,
teardown):
"""
Node requests PRE-PREPARE only once after getting PREPAREs.
"""
slow_node, other_nodes, primary_node, \
other_primary_nodes = split_nodes(txnPoolNodeSet)
# Drop PrePrepares and Prepares
slow_node.nodeIbStasher.delay(ppDelay(300, 0))
slow_node.nodeIbStasher.delay(pDelay(300, 0))
sdk_send_batches_of_random_and_check(looper,
txnPoolNodeSet,
sdk_pool_handle,
sdk_wallet_client,
num_reqs=10,
num_batches=5)
slow_node.nodeIbStasher.drop_delayeds()
slow_node.nodeIbStasher.resetDelays()
old_count_req = count_requested_preprepare_req(slow_node)
old_count_resp = count_requested_preprepare_resp(slow_node)
def chk(increase=True):
# Method is called
assert count_requested_preprepare_req(slow_node) > old_count_req
# Requesting Preprepare
assert count_requested_preprepare_resp(
slow_node) - old_count_resp == (1 if increase else 0)
for pp in primary_node.master_replica.sentPrePrepares.values():
for rep in [n.master_replica for n in other_primary_nodes]:
prepare = Prepare(rep.instId,
pp.viewNo,
pp.ppSeqNo,
pp.ppTime,
pp.digest,
pp.stateRootHash,
pp.txnRootHash
)
rep.send(prepare)
looper.run(eventually(chk, True, retryWait=1))
old_count_resp = count_requested_preprepare_resp(slow_node)
prepare = Prepare(rep.instId,
pp.viewNo,
pp.ppSeqNo,
pp.ppTime,
pp.digest,
pp.stateRootHash,
pp.txnRootHash
)
rep.send(prepare)
looper.run(eventually(chk, False, retryWait=1))
old_count_req = count_requested_preprepare_req(slow_node)
old_count_resp = count_requested_preprepare_resp(slow_node)
def test_non_primary_recvs_3phase_message_outside_watermarks(
chkFreqPatched, reqs_for_logsize, looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_client):
"""
A node is slow in receiving PRE-PREPAREs and PREPAREs. A lot of requests
are sent and the slow node has started receiving COMMITs outside of its
watermarks and so stashes them. Also this node is slow in receiving
CHECKPOINTs. So a catch-up does not occur on it.
Then the slow node eventually receives the sent PRE-PREPAREs and PREPAREs
and so orders the 3PC-batches between its watermarks. The other nodes
discard the COMMITs from the slow node since they have already achieved
stable checkpoints for these COMMITs.
After that the slow node eventually receives the sent CHECKPOINTs from
the other nodes and so stabilizes own completed checkpoints and updates its
watermarks. A catch-up is not triggered because no received checkpoints are
stashed. Since now the watermarks have been updated, the slow node
processes 3PC-messages stashed earlier and its ledger becomes equal to the
ledgers of the other nodes.
"""
backupInstId = 1
npr = getNonPrimaryReplicas(txnPoolNodeSet, backupInstId)
slowReplica = npr[0]
slowNode = slowReplica.node
slowNode.nodeIbStasher.delay(ppDelay(300, backupInstId))
slowNode.nodeIbStasher.delay(pDelay(300, backupInstId))
slowNode.nodeIbStasher.delay(chk_delay(300))
initialDomainLedgerSize = slowNode.domainLedger.size
oldStashCount = slowReplica.spylog.count(TestReplica.stashOutsideWatermarks.__name__)
# 1. Send requests more than fit between the watermarks on the slow node
sdk_send_random_and_check(looper, txnPoolNodeSet, sdk_pool_handle, sdk_wallet_client, reqs_for_logsize + 2)
# Verify that the slow node stashes the batches outside of its watermarks
newStashCount = slowReplica.spylog.count(TestReplica.stashOutsideWatermarks.__name__)
assert newStashCount > oldStashCount
oldDiscardCounts = discardCounts([n.replicas[backupInstId] for n in txnPoolNodeSet if n != slowNode],
'achieved stable checkpoint')
# 2. Deliver the sent PREPREPAREs and PREPAREs to the slow node
slowNode.nodeIbStasher.reset_delays_and_process_delayeds(PREPREPARE, PREPARE)
# Verify that the slow node orders the 3PC-batches between its watermarks
# but no more.
looper.runFor(waits.expectedTransactionExecutionTime(len(txnPoolNodeSet)))
checkNodeDataForInequality(slowNode, *[n for n in txnPoolNodeSet if n != slowNode])
assert slowNode.domainLedger.size - initialDomainLedgerSize == reqs_for_logsize
# Also verify that the other nodes discard the COMMITs from the slow node
# since they have already achieved stable checkpoints for these COMMITs.
counts = discardCounts(
[n.replicas[backupInstId] for n in txnPoolNodeSet if n != slowNode],
'achieved stable checkpoint')
for nm, count in counts.items():
assert count > oldDiscardCounts[nm]
oldCatchupTimes = slowNode.spylog.count(Node.start_catchup)
# 3. Deliver the sent CHECKPOINTs to the slow node
slowNode.nodeIbStasher.reset_delays_and_process_delayeds(CHECKPOINT)
# Verify that the slow node processes 3PC-messages stashed earlier and its
# ledger becomes equal to the ledgers of the other nodes while a catch-up
# is not made.
waitNodeDataEquality(looper, slowNode, *[n for n in txnPoolNodeSet if n != slowNode])
assert slowNode.domainLedger.size - initialDomainLedgerSize == reqs_for_logsize + 2
newCatchupTimes = slowNode.spylog.count(Node.start_catchup)
assert newCatchupTimes == oldCatchupTimes
