void replica::on_group_check(const group_check_request& request, __out_param group_check_response& response)
{
ddebug(
"%s: on_group_check from %s:%d",
name(), request.config.primary.name.c_str(), request.config.primary.port
);
if (request.config.ballot < get_ballot())
{
response.err = ERR_VERSION_OUTDATED;
return;
}
else if (request.config.ballot > get_ballot())
{
update_local_configuration(request.config);
}
else if (is_same_ballot_status_change_allowed(status(), request.config.status))
{
update_local_configuration(request.config, true);
}
switch (status())
{
case PS_INACTIVE:
break;
case PS_SECONDARY:
if (request.last_committed_decree > last_committed_decree())
{
_prepare_list->commit(request.last_committed_decree, true);
}
break;
case PS_POTENTIAL_SECONDARY:
init_learn(request.learner_signature);
break;
case PS_ERROR:
break;
default:
dassert (false, "");
}
response.gpid = get_gpid();
response.node = primary_address();
response.err = ERR_SUCCESS;
if (status() == PS_ERROR)
{
response.err = ERR_INVALID_STATE;
}
response.last_committed_decree_in_app = _app->last_committed_decree();
response.last_committed_decree_in_prepare_list = last_committed_decree();
response.learner_status_ = _potential_secondary_states.learning_status;
response.learner_signature = _potential_secondary_states.learning_signature;
}
// from primary
void replica::on_remove(const replica_configuration& request)
{
if (request.ballot < get_ballot())
return;
dassert (request.status == PS_INACTIVE, "");
update_local_configuration(request);
}
bool replica::update_local_configuration_with_no_ballot_change(partition_status s)
{
if (status() == s)
return false;
auto config = _config;
config.status = s;
return update_local_configuration(config, true);
}
// @ secondary
void replica::on_copy_checkpoint(const replica_configuration& request, /*out*/ learn_response& response)
{
check_hashed_access();
if (request.ballot > get_ballot())
{
if (!update_local_configuration(request))
{
response.err = ERR_INVALID_STATE;
return;
}
}
if (status() != PS_SECONDARY)
{
response.err = ERR_INVALID_STATE;
return;
}
if (_app->last_durable_decree() == 0)
{
response.err = ERR_OBJECT_NOT_FOUND;
return;
}
blob placeholder;
int err = _app->get_checkpoint(0, placeholder, response.state);
if (err != 0)
{
response.err = ERR_LEARN_FILE_FAILED;
}
else
{
response.err = ERR_OK;
response.last_committed_decree = last_committed_decree();
response.base_local_dir = _app->data_dir();
response.address = _stub->_primary_address;
}
}
bool replica::update_configuration(const partition_configuration& config)
{
dassert (config.ballot >= get_ballot(), "");
replica_configuration rconfig;
replica_helper::get_replica_config(config, primary_address(), rconfig);
if (rconfig.status == PS_PRIMARY &&
(rconfig.ballot > get_ballot() || status() != PS_PRIMARY)
)
{
_primary_states.reset_membership(config, config.primary != primary_address());
}
if (config.ballot > get_ballot() ||
is_same_ballot_status_change_allowed(status(), rconfig.status)
)
{
return update_local_configuration(rconfig, true);
}
else
return false;
}
void replica::on_group_check(const group_check_request& request, /*out*/ group_check_response& response)
{
check_hashed_access();
ddebug(
"%s: process group check, primary = %s, ballot = %" PRId64 ", status = %s, last_committed_decree = %" PRId64,
name(), request.config.primary.to_string(),
request.config.ballot, enum_to_string(request.config.status),
request.last_committed_decree
);
if (request.config.ballot < get_ballot())
{
response.err = ERR_VERSION_OUTDATED;
dwarn("%s: on_group_check reply %s", name(), response.err.to_string());
return;
}
else if (request.config.ballot > get_ballot())
{
if (!update_local_configuration(request.config))
{
response.err = ERR_INVALID_STATE;
dwarn("%s: on_group_check reply %s", name(), response.err.to_string());
return;
}
}
else if (is_same_ballot_status_change_allowed(status(), request.config.status))
{
update_local_configuration(request.config, true);
}
switch (status())
{
case partition_status::PS_INACTIVE:
break;
case partition_status::PS_SECONDARY:
if (request.last_committed_decree > last_committed_decree())
{
_prepare_list->commit(request.last_committed_decree, COMMIT_TO_DECREE_HARD);
}
break;
case partition_status::PS_POTENTIAL_SECONDARY:
init_learn(request.config.learner_signature);
break;
case partition_status::PS_ERROR:
break;
default:
dassert (false, "");
}
response.pid = get_gpid();
response.node = _stub->_primary_address;
response.err = ERR_OK;
if (status() == partition_status::PS_ERROR)
{
response.err = ERR_INVALID_STATE;
dwarn("%s: on_group_check reply %s", name(), response.err.to_string());
}
response.last_committed_decree_in_app = _app->last_committed_decree();
response.last_committed_decree_in_prepare_list = last_committed_decree();
response.learner_status_ = _potential_secondary_states.learning_status;
response.learner_signature = _potential_secondary_states.learning_version;
}
void replica::on_prepare(dsn_message_t request)
{
check_hashed_access();
replica_configuration rconfig;
mutation_ptr mu;
{
rpc_read_stream reader(request);
unmarshall(reader, rconfig);
mu = mutation::read_from(reader, request);
}
decree decree = mu->data.header.decree;
dinfo("%s: mutation %s on_prepare", name(), mu->name());
dassert(mu->data.header.ballot == rconfig.ballot, "");
if (mu->data.header.ballot < get_ballot())
{
derror("%s: mutation %s on_prepare skipped due to old view", name(), mu->name());
// no need response because the rpc should have been cancelled on primary in this case
return;
}
// update configuration when necessary
else if (rconfig.ballot > get_ballot())
{
if (!update_local_configuration(rconfig))
{
derror(
"%s: mutation %s on_prepare failed as update local configuration failed, state = %s",
name(), mu->name(),
enum_to_string(status())
);
ack_prepare_message(ERR_INVALID_STATE, mu);
return;
}
}
if (PS_INACTIVE == status() || PS_ERROR == status())
{
derror(
"%s: mutation %s on_prepare failed as invalid replica state, state = %s",
name(), mu->name(),
enum_to_string(status())
);
ack_prepare_message(
(PS_INACTIVE == status() && _inactive_is_transient) ? ERR_INACTIVE_STATE : ERR_INVALID_STATE,
mu
);
return;
}
else if (PS_POTENTIAL_SECONDARY == status())
{
// new learning process
if (rconfig.learner_signature != _potential_secondary_states.learning_signature)
{
init_learn(rconfig.learner_signature);
// no need response as rpc is already gone
return;
}
if (!(_potential_secondary_states.learning_status == LearningWithPrepare
|| _potential_secondary_states.learning_status == LearningSucceeded))
{
derror(
"%s: mutation %s on_prepare skipped as invalid learning status, state = %s, learning_status = %s",
name(), mu->name(),
enum_to_string(status()),
enum_to_string(_potential_secondary_states.learning_status)
);
// no need response as rpc is already gone
return;
}
}
dassert (rconfig.status == status(), "");
if (decree <= last_committed_decree())
{
ack_prepare_message(ERR_OK, mu);
return;
}
// real prepare start
auto mu2 = _prepare_list->get_mutation_by_decree(decree);
if (mu2 != nullptr && mu2->data.header.ballot == mu->data.header.ballot)
{
if (mu2->is_logged())
{
ack_prepare_message(ERR_OK, mu);
}
else
{
derror("%s: mutation %s on_prepare skipped as it is duplicate", name(), mu->name());
// response will be unnecessary when we add retry logic in rpc engine.
// the retried rpc will use the same id therefore it will be considered responsed
// even the response is for a previous try.
}
return;
}
error_code err = _prepare_list->prepare(mu, status());
dassert (err == ERR_OK, "");
if (PS_POTENTIAL_SECONDARY == status())
{
dassert (mu->data.header.decree <= last_committed_decree() + _options->max_mutation_count_in_prepare_list, "");
}
else
{
dassert (PS_SECONDARY == status(), "");
dassert (mu->data.header.decree <= last_committed_decree() + _options->staleness_for_commit, "");
}
dassert(mu->log_task() == nullptr, "");
mu->log_task() = _stub->_log->append(mu,
LPC_WRITE_REPLICATION_LOG,
this,
std::bind(&replica::on_append_log_completed, this, mu,
std::placeholders::_1,
std::placeholders::_2),
gpid_to_hash(get_gpid())
);
}
void replica::on_prepare(dsn_message_t request)
{
check_hashed_access();
replica_configuration rconfig;
mutation_ptr mu;
{
msg_binary_reader reader(request);
unmarshall(reader, rconfig);
mu = mutation::read_from(reader, request);
}
decree decree = mu->data.header.decree;
ddebug( "%s: mutation %s on_prepare", name(), mu->name());
dassert (mu->data.header.ballot == rconfig.ballot, "");
if (mu->data.header.ballot < get_ballot())
{
ddebug( "%s: mutation %s on_prepare skipped due to old view", name(), mu->name());
return;
}
// update configuration when necessary
else if (rconfig.ballot > get_ballot())
{
if (!update_local_configuration(rconfig))
{
ddebug(
"%s: mutation %s on_prepare to %s failed as update local configuration failed",
name(), mu->name(),
enum_to_string(status())
);
ack_prepare_message(ERR_INVALID_STATE, mu);
return;
}
}
if (PS_INACTIVE == status() || PS_ERROR == status())
{
ddebug(
"%s: mutation %s on_prepare to %s skipped",
name(), mu->name(),
enum_to_string(status())
);
ack_prepare_message(
(PS_INACTIVE == status() && _inactive_is_transient) ? ERR_INACTIVE_STATE : ERR_INVALID_STATE,
mu
);
return;
}
else if (PS_POTENTIAL_SECONDARY == status())
{
if (_potential_secondary_states.learning_status != LearningWithPrepare && _potential_secondary_states.learning_status != LearningSucceeded)
{
ddebug(
"%s: mutation %s on_prepare to %s skipped, learnings state = %s",
name(), mu->name(),
enum_to_string(status()),
enum_to_string(_potential_secondary_states.learning_status)
);
// do not retry as there may retries later
return;
}
}
dassert (rconfig.status == status(), "");
if (decree <= last_committed_decree())
{
ack_prepare_message(ERR_OK, mu);
return;
}
// real prepare start
auto mu2 = _prepare_list->get_mutation_by_decree(decree);
if (mu2 != nullptr && mu2->data.header.ballot == mu->data.header.ballot)
{
ddebug( "%s: mutation %s redundant prepare skipped", name(), mu->name());
if (mu2->is_logged() || _options.prepare_ack_on_secondary_before_logging_allowed)
{
ack_prepare_message(ERR_OK, mu);
}
return;
}
error_code err = _prepare_list->prepare(mu, status());
dassert (err == ERR_OK, "");
if (PS_POTENTIAL_SECONDARY == status())
{
dassert (mu->data.header.decree <= last_committed_decree() + _options.staleness_for_start_prepare_for_potential_secondary, "");
}
else
{
dassert (PS_SECONDARY == status(), "");
dassert (mu->data.header.decree <= last_committed_decree() + _options.staleness_for_commit, "");
}
// ack without logging
if (_options.prepare_ack_on_secondary_before_logging_allowed)
{
ack_prepare_message(err, mu);
}
// write log
dassert (mu->log_task() == nullptr, "");
mu->log_task() = _stub->_log->append(mu,
LPC_WRITE_REPLICATION_LOG,
this,
std::bind(&replica::on_append_log_completed, this, mu, std::placeholders::_1, std::placeholders::_2),
gpid_to_hash(get_gpid())
);
dassert(mu->log_task() != nullptr, "");
}
