/*
- * Copyright (c) 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2012, 2013 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <config.h>
#include "reconnect.h"
-#include <assert.h>
#include <stdlib.h>
-#include "poll-loop.h"
-#include "vlog.h"
+#include "openvswitch/poll-loop.h"
+#include "util.h"
+#include "openvswitch/vlog.h"
VLOG_DEFINE_THIS_MODULE(reconnect);
bool passive;
enum vlog_level info; /* Used for informational messages. */
- uint8_t dscp;
-
/* State. */
enum state state;
long long int state_entered;
int backoff;
- long long int last_received;
+ long long int last_activity;
long long int last_connected;
long long int last_disconnected;
unsigned int max_tries;
+ unsigned int backoff_free_tries;
/* These values are simply for statistics reporting, not otherwise used
* directly by anything internal. */
fsm->state = S_VOID;
fsm->state_entered = now;
fsm->backoff = 0;
- fsm->last_received = now;
+ fsm->last_activity = now;
fsm->last_connected = LLONG_MAX;
fsm->last_disconnected = LLONG_MAX;
fsm->max_tries = UINT_MAX;
/* Returns the "probe interval" for 'fsm' in milliseconds. If this is zero, it
* disables the connection keepalive feature. If it is nonzero, then if the
- * interval passes while 'fsm' is connected and without reconnect_received()
+ * interval passes while 'fsm' is connected and without reconnect_activity()
* being called for 'fsm', reconnect_run() returns RECONNECT_PROBE. If the
- * interval passes again without reconnect_received() being called,
+ * interval passes again without reconnect_activity() being called,
* reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'. */
int
reconnect_get_probe_interval(const struct reconnect *fsm)
return fsm->probe_interval;
}
-/* Returns the dscp value used for establishing the connection between the
- * manager and the database. */
-uint8_t
-reconnect_get_dscp(const struct reconnect *fsm)
-{
- return fsm->dscp;
-}
-
/* Limits the maximum number of times that 'fsm' will ask the client to try to
* reconnect to 'max_tries'. UINT_MAX (the default) means an unlimited number
* of tries.
return fsm->max_tries;
}
+/* Sets the number of connection attempts that will be made without backoff to
+ * 'backoff_free_tries'. Values 0 and 1 both represent a single attempt. */
+void
+reconnect_set_backoff_free_tries(struct reconnect *fsm,
+ unsigned int backoff_free_tries)
+{
+ fsm->backoff_free_tries = backoff_free_tries;
+}
+
/* Configures the backoff parameters for 'fsm'. 'min_backoff' is the minimum
* number of milliseconds, and 'max_backoff' is the maximum, between connection
- * attempts.
+ * attempts. The current backoff is also the duration that 'fsm' is willing to
+ * wait for a given connection to succeed or fail.
*
* 'min_backoff' must be at least 1000, and 'max_backoff' must be greater than
* or equal to 'min_backoff'.
/* Sets the "probe interval" for 'fsm' to 'probe_interval', in milliseconds.
* If this is zero, it disables the connection keepalive feature. If it is
* nonzero, then if the interval passes while 'fsm' is connected and without
- * reconnect_received() being called for 'fsm', reconnect_run() returns
- * RECONNECT_PROBE. If the interval passes again without reconnect_received()
+ * reconnect_activity() being called for 'fsm', reconnect_run() returns
+ * RECONNECT_PROBE. If the interval passes again without reconnect_activity()
* being called, reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'.
*
* If 'probe_interval' is nonzero, then it will be forced to a value of at
fsm->probe_interval = probe_interval ? MAX(1000, probe_interval) : 0;
}
-/* Sets the dscp value to be used for establishing a connection between the
- * manager and the database. */
-void
-reconnect_set_dscp(struct reconnect *fsm, uint8_t dscp)
-{
- fsm->dscp = dscp;
-}
-
/* Returns true if 'fsm' is in passive mode, false if 'fsm' is in active mode
* (the default). */
bool
if (fsm->state & (S_ACTIVE | S_IDLE)) {
if (error > 0) {
VLOG_WARN("%s: connection dropped (%s)",
- fsm->name, strerror(error));
+ fsm->name, ovs_strerror(error));
} else if (error == EOF) {
VLOG(fsm->info, "%s: connection closed by peer", fsm->name);
} else {
} else if (fsm->state == S_LISTENING) {
if (error > 0) {
VLOG_WARN("%s: error listening for connections (%s)",
- fsm->name, strerror(error));
+ fsm->name, ovs_strerror(error));
} else {
VLOG(fsm->info, "%s: error listening for connections",
fsm->name);
}
- } else {
+ } else if (fsm->backoff < fsm->max_backoff) {
const char *type = fsm->passive ? "listen" : "connection";
if (error > 0) {
- VLOG_WARN("%s: %s attempt failed (%s)",
- fsm->name, type, strerror(error));
+ VLOG_INFO("%s: %s attempt failed (%s)",
+ fsm->name, type, ovs_strerror(error));
} else {
VLOG(fsm->info, "%s: %s attempt timed out", fsm->name, type);
}
+ } else {
+ /* We have reached the maximum backoff, so suppress logging to
+ * avoid wastefully filling the log. (Previously we logged that we
+ * were suppressing further logging, see below.) */
}
if (fsm->state & (S_ACTIVE | S_IDLE)) {
fsm->last_disconnected = now;
}
+
+ if (!reconnect_may_retry(fsm)) {
+ reconnect_transition__(fsm, now, S_VOID);
+ return;
+ }
+
/* Back off. */
- if (fsm->state & (S_ACTIVE | S_IDLE)
- && (fsm->last_received - fsm->last_connected >= fsm->backoff
- || fsm->passive)) {
+ if (fsm->backoff_free_tries > 1) {
+ fsm->backoff_free_tries--;
+ fsm->backoff = 0;
+ } else if (fsm->state & (S_ACTIVE | S_IDLE)
+ && (fsm->last_activity - fsm->last_connected >= fsm->backoff
+ || fsm->passive)) {
fsm->backoff = fsm->passive ? 0 : fsm->min_backoff;
} else {
if (fsm->backoff < fsm->min_backoff) {
fsm->backoff = fsm->min_backoff;
- } else if (fsm->backoff >= fsm->max_backoff / 2) {
- fsm->backoff = fsm->max_backoff;
- } else {
+ } else if (fsm->backoff < fsm->max_backoff / 2) {
fsm->backoff *= 2;
- }
- if (fsm->passive) {
- VLOG(fsm->info, "%s: waiting %.3g seconds before trying to "
- "listen again", fsm->name, fsm->backoff / 1000.0);
+ VLOG(fsm->info, "%s: waiting %.3g seconds before %s",
+ fsm->name, fsm->backoff / 1000.0,
+ fsm->passive ? "trying to listen again" : "reconnect");
} else {
- VLOG(fsm->info, "%s: waiting %.3g seconds before reconnect",
- fsm->name, fsm->backoff / 1000.0);
+ if (fsm->backoff < fsm->max_backoff) {
+ VLOG_INFO("%s: continuing to %s in the background but "
+ "suppressing further logging", fsm->name,
+ fsm->passive ? "try to listen" : "reconnect");
+ }
+ fsm->backoff = fsm->max_backoff;
}
}
-
- reconnect_transition__(fsm, now,
- reconnect_may_retry(fsm) ? S_BACKOFF : S_VOID);
+ reconnect_transition__(fsm, now, S_BACKOFF);
}
}
if (fsm->state != S_CONNECTING) {
if (fsm->passive) {
VLOG(fsm->info, "%s: listening...", fsm->name);
- } else {
+ } else if (fsm->backoff < fsm->max_backoff) {
VLOG(fsm->info, "%s: connecting...", fsm->name);
}
reconnect_transition__(fsm, now, S_CONNECTING);
/* Tell 'fsm' that the connection was successful.
*
* The FSM will start the probe interval timer, which is reset by
- * reconnect_received(). If the timer expires, a probe will be sent (by
+ * reconnect_activity(). If the timer expires, a probe will be sent (by
* returning RECONNECT_PROBE from reconnect_run()). If the timer expires
* again without being reset, the connection will be aborted (by returning
* RECONNECT_DISCONNECT from reconnect_run()). */
reconnect_disconnected(fsm, now, error);
}
-/* Tell 'fsm' that some data was received. This resets the probe interval
- * timer, so that the connection is known not to be idle. */
+/* Tell 'fsm' that some activity has occurred on the connection. This resets
+ * the probe interval timer, so that the connection is known not to be idle. */
void
-reconnect_received(struct reconnect *fsm, long long int now)
+reconnect_activity(struct reconnect *fsm, long long int now)
{
- if (fsm->state != S_ACTIVE) {
+ if (fsm->state == S_IDLE) {
reconnect_transition__(fsm, now, S_ACTIVE);
}
- fsm->last_received = now;
+ fsm->last_activity = now;
}
static void
static long long int
reconnect_deadline__(const struct reconnect *fsm)
{
- assert(fsm->state_entered != LLONG_MIN);
+ ovs_assert(fsm->state_entered != LLONG_MIN);
switch (fsm->state) {
case S_VOID:
case S_LISTENING:
case S_ACTIVE:
if (fsm->probe_interval) {
- long long int base = MAX(fsm->last_received, fsm->state_entered);
+ long long int base = MAX(fsm->last_activity, fsm->state_entered);
return base + fsm->probe_interval;
}
return LLONG_MAX;
case S_IDLE:
- return fsm->state_entered + fsm->probe_interval;
+ if (fsm->probe_interval) {
+ return fsm->state_entered + fsm->probe_interval;
+ }
+ return LLONG_MAX;
case S_RECONNECT:
return fsm->state_entered;
}
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
/* Assesses whether any action should be taken on 'fsm'. The return value is
case S_ACTIVE:
VLOG_DBG("%s: idle %lld ms, sending inactivity probe", fsm->name,
- now - MAX(fsm->last_received, fsm->state_entered));
+ now - MAX(fsm->last_activity, fsm->state_entered));
reconnect_transition__(fsm, now, S_IDLE);
return RECONNECT_PROBE;
return 0;
}
- NOT_REACHED();
+ OVS_NOT_REACHED();
} else {
return 0;
}
struct reconnect_stats *stats)
{
stats->creation_time = fsm->creation_time;
- stats->last_received = fsm->last_received;
+ stats->last_activity = fsm->last_activity;
stats->last_connected = fsm->last_connected;
stats->last_disconnected = fsm->last_disconnected;
stats->backoff = fsm->backoff;
projects / mirror_ovs.git / blobdiff