Skip to content

Commit

Permalink
Make Agent Register request asynchronous
Browse files Browse the repository at this point in the history 
The "Register" agent->controller method call could get stuck
when the IP address configured as ControllerHost does not exist.
This timeout of about 2 seconds could completely occupy the event
loop and would not allow other calls on the DBus API of the Agent.

Making the "Register" method call asynchronous we allow the event
loop to stay unoccupied and be able to accept SwitchController
request.

Resolves: eclipse-bluechi#950
Signed-off-by: Mark Kemel <[email protected]>
  • Loading branch information
@mkemel
mkemel committed Oct 30, 2024
1 parent 214c8fb commit 7027c07
Show file tree
Hide file tree
Showing 3 changed files with 103 additions and 44 deletions.
139 changes: 97 additions & 42 deletions src/agent/agent.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -110,6 +110,7 @@ char *agent_is_online(Agent *agent) {
static int agent_stop_local_proxy_service(Agent *agent, ProxyService *proxy);
static bool agent_connect(Agent *agent);
static bool agent_reconnect(Agent *agent);
static void agent_disconnect(Agent *agent);

static int agent_disconnected(UNUSED sd_bus_message *message, void *userdata, UNUSED sd_bus_error *error) {
Agent *agent = (Agent *) userdata;
Expand Down Expand Up @@ -169,6 +170,17 @@ static int agent_heartbeat_timer_callback(sd_event_source *event_source, UNUSED
Agent *agent = (Agent *) userdata;

int r = 0;
/* Being in CONNECTING state when the timer callback is executed implies that
* the agent hasn't received a reply from the controller yet. In this case,
* we drop the existing register call and the agent is set into RETRY state.*/
if (agent->connection_state == AGENT_CONNECTION_STATE_CONNECTING) {
bc_log_error("Agent connection attempt failed, retrying");
if (agent->register_call_slot != NULL) {
sd_bus_slot_unrefp(&agent->register_call_slot);
agent->register_call_slot = NULL;
}
agent->connection_state = AGENT_CONNECTION_STATE_RETRY;
}
if (agent->connection_state == AGENT_CONNECTION_STATE_CONNECTED &&
agent_check_controller_liveness(agent)) {
r = sd_bus_emit_signal(
Expand Down Expand Up @@ -532,6 +544,9 @@ void agent_unref(Agent *agent) {
sd_event_unrefp(&agent->event);
}

if (agent->register_call_slot != NULL) {
sd_bus_slot_unrefp(&agent->register_call_slot);
}
if (agent->metrics_slot != NULL) {
sd_bus_slot_unrefp(&agent->metrics_slot);
}
Expand Down Expand Up @@ -2680,6 +2695,7 @@ void agent_stop(Agent *agent) {

bc_log_debug("Stopping agent");

agent_disconnect(agent);
agent->connection_state = AGENT_CONNECTION_STATE_DISCONNECTED;
int r = sd_bus_emit_properties_changed(
agent->api_bus, BC_AGENT_OBJECT_PATH, AGENT_INTERFACE, "Status", NULL);
Expand All @@ -2694,51 +2710,18 @@ void agent_stop(Agent *agent) {
}
}

static bool agent_connect(Agent *agent) {
peer_bus_close(agent->peer_dbus);

bc_log_infof("Connecting to controller on %s", agent->assembled_controller_address);

agent->peer_dbus = peer_bus_open(
agent->event, "peer-bus-to-controller", agent->assembled_controller_address);
if (agent->peer_dbus == NULL) {
bc_log_error("Failed to open peer dbus");
return false;
}

bus_socket_set_options(agent->peer_dbus, agent->peer_socket_options);

int r = sd_bus_add_object_vtable(
agent->peer_dbus,
NULL,
INTERNAL_AGENT_OBJECT_PATH,
INTERNAL_AGENT_INTERFACE,
internal_agent_vtable,
agent);
if (r < 0) {
bc_log_errorf("Failed to add agent vtable: %s", strerror(-r));
return false;
}
static bool agent_process_register_callback(sd_bus_message *m, Agent *agent) {
int r = 0;

_cleanup_sd_bus_message_ sd_bus_message *bus_msg = NULL;
sd_bus_error error = SD_BUS_ERROR_NULL;
r = sd_bus_call_method(
agent->peer_dbus,
BC_DBUS_NAME,
INTERNAL_CONTROLLER_OBJECT_PATH,
INTERNAL_CONTROLLER_INTERFACE,
"Register",
&error,
&bus_msg,
"s",
agent->name);
if (r < 0) {
bc_log_errorf("Registering as '%s' failed: %s", agent->name, error.message);
sd_bus_error_free(&error);
return false;
if (sd_bus_message_is_method_error(m, NULL)) {
bc_log_errorf("Registering as '%s' failed: %s",
agent->name,
sd_bus_message_get_error(m)->message);
return -EPERM;
}

r = sd_bus_message_read(bus_msg, "");
bc_log_info("Register call response received");
r = sd_bus_message_read(m, "");
if (r < 0) {
bc_log_errorf("Failed to parse response message: %s", strerror(-r));
return false;
Expand Down Expand Up @@ -2805,6 +2788,68 @@ static bool agent_connect(Agent *agent) {
return true;
}

static int agent_register_callback(sd_bus_message *m, void *userdata, UNUSED sd_bus_error *ret_error) {
Agent *agent = (Agent *) userdata;

if (agent->connection_state != AGENT_CONNECTION_STATE_CONNECTING) {
bc_log_error("Agent is not in CONNECTING state, dropping Register callback");
sd_bus_slot_unrefp(&agent->register_call_slot);
agent->register_call_slot = NULL;
return 0;
}

if (!agent_process_register_callback(m, agent)) {
agent->connection_state = AGENT_CONNECTION_STATE_RETRY;
}

return 0;
}

static bool agent_connect(Agent *agent) {
bc_log_infof("Connecting to controller on %s", agent->assembled_controller_address);
agent->connection_state = AGENT_CONNECTION_STATE_CONNECTING;

agent->peer_dbus = peer_bus_open(
agent->event, "peer-bus-to-controller", agent->assembled_controller_address);
if (agent->peer_dbus == NULL) {
bc_log_error("Failed to open peer dbus");
return false;
}

bus_socket_set_options(agent->peer_dbus, agent->peer_socket_options);

int r = sd_bus_add_object_vtable(
agent->peer_dbus,
NULL,
INTERNAL_AGENT_OBJECT_PATH,
INTERNAL_AGENT_INTERFACE,
internal_agent_vtable,
agent);
if (r < 0) {
bc_log_errorf("Failed to add agent vtable: %s", strerror(-r));
return false;
}

_cleanup_sd_bus_message_ sd_bus_message *bus_msg = NULL;
r = sd_bus_call_method_async(
agent->peer_dbus,
&agent->register_call_slot,
BC_DBUS_NAME,
INTERNAL_CONTROLLER_OBJECT_PATH,
INTERNAL_CONTROLLER_INTERFACE,
"Register",
agent_register_callback,
agent,
"s",
agent->name);
if (r < 0) {
bc_log_errorf("Registering as '%s' failed: %s", agent->name, strerror(-r));
return false;
}

return true;
}

static bool agent_reconnect(Agent *agent) {
_cleanup_free_ char *assembled_controller_address = NULL;

Expand All @@ -2827,9 +2872,19 @@ static bool agent_reconnect(Agent *agent) {
}
}

agent_disconnect(agent);
return agent_connect(agent);
}

static void agent_disconnect(Agent *agent) {
if (agent->register_call_slot != NULL) {
sd_bus_slot_unref(agent->register_call_slot);
agent->register_call_slot = NULL;
}
peer_bus_close(agent->peer_dbus);
agent->peer_dbus = NULL;
}

static int stop_proxy_callback(sd_bus_message *m, void *userdata, UNUSED sd_bus_error *ret_error) {
UNUSED _cleanup_systemd_request_ SystemdRequest *req = userdata;

Expand Down
5 changes: 4 additions & 1 deletion src/agent/agent.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,8 @@ typedef struct JobTracker JobTracker;
typedef enum {
AGENT_CONNECTION_STATE_DISCONNECTED,
AGENT_CONNECTION_STATE_CONNECTED,
AGENT_CONNECTION_STATE_RETRY
AGENT_CONNECTION_STATE_RETRY,
AGENT_CONNECTION_STATE_CONNECTING
} AgentConnectionState;

struct Agent {
Expand Down Expand Up @@ -81,6 +82,8 @@ struct Agent {
sd_bus *systemd_dbus;
sd_bus *peer_dbus;

sd_bus_slot *register_call_slot;

bool metrics_enabled;
sd_bus_slot *metrics_slot;

Expand Down
3 changes: 2 additions & 1 deletion src/libbluechi/bus/bus.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -77,13 +77,14 @@ sd_bus *peer_bus_open(sd_event *event, const char *dbus_description, const char

int peer_bus_close(sd_bus *peer_dbus) {
if (peer_dbus != NULL) {
bc_log_debug("Closing peer bus");
int r = sd_bus_detach_event(peer_dbus);
if (r < 0) {
bc_log_errorf("Failed to detach bus from event: %s", strerror(-r));
return r;
}

sd_bus_flush_close_unref(peer_dbus);
sd_bus_close_unref(peer_dbus);
}

return 0;
Expand Down

0 comments on commit 7027c07

Please sign in to comment.