#include <urcu/list.h>
#include <signal.h>
+#include <bin/lttng-consumerd/health-consumerd.h>
#include <common/common.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <common/relayd/relayd.h>
cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
send_node) {
+
+ health_code_update();
+
cds_list_del(&stream->send_node);
ustctl_destroy_stream(stream->ustream);
free(stream);
int wait_fd;
int ust_metadata_pipe[2];
+ health_code_update();
+
if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
if (ret < 0) {
struct lttng_consumer_channel *channel,
struct lttng_consumer_local_data *ctx, int *relayd_error)
{
- int ret, ret_code = LTTNG_OK;
+ int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
struct lttng_consumer_stream *stream;
+ uint64_t net_seq_idx = -1ULL;
assert(channel);
assert(ctx);
if (channel->relayd_id != (uint64_t) -1ULL) {
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+
+ health_code_update();
+
/* Try to send the stream to the relayd if one is available. */
ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
if (ret < 0) {
}
ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
}
+ if (net_seq_idx == -1ULL) {
+ net_seq_idx = stream->net_seq_idx;
+ }
+ }
+ ret = consumer_send_relayd_streams_sent(net_seq_idx);
+ if (ret < 0) {
+ /*
+ * Flag that the relayd was the problem here probably due to a
+ * communicaton error on the socket.
+ */
+ if (relayd_error) {
+ *relayd_error = 1;
+ }
+ ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
}
}
/* Inform sessiond that we are about to send channel and streams. */
ret = consumer_send_status_msg(sock, ret_code);
- if (ret < 0 || ret_code != LTTNG_OK) {
+ if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
/*
* Either the session daemon is not responding or the relayd died so we
* stop now.
/* The channel was sent successfully to the sessiond at this point. */
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+
+ health_code_update();
+
/* Send stream to session daemon. */
ret = send_sessiond_stream(sock, stream);
if (ret < 0) {
return 0;
error:
- if (ret_code != LTTNG_OK) {
+ if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
ret = -1;
}
return ret;
/* Send streams to the corresponding thread. */
cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
send_node) {
+
+ health_code_update();
+
/* Sending the stream to the thread. */
ret = send_stream_to_thread(stream, ctx);
if (ret < 0) {
cds_lfht_for_each_entry_duplicate(ht->ht,
ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
&channel->key, &iter.iter, stream, node_channel_id.node) {
+
+ health_code_update();
+
ustctl_flush_buffer(stream->ustream, 1);
}
error:
*/
static int _close_metadata(struct lttng_consumer_channel *channel)
{
- int ret = LTTNG_OK;
+ int ret = LTTCOMM_CONSUMERD_SUCCESS;
assert(channel);
assert(channel->type == CONSUMER_CHANNEL_TYPE_METADATA);
}
assert(!metadata_channel->monitor);
+ health_code_update();
+
/*
* Ask the sessiond if we have new metadata waiting and update the
* consumer metadata cache.
goto error;
}
+ health_code_update();
+
/*
* The metadata stream is NOT created in no monitor mode when the channel
* is created on a sessiond ask channel command.
}
do {
+ health_code_update();
+
ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
if (ret < 0) {
goto error_stream;
DBG("UST consumer snapshot channel %" PRIu64, key);
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+
+ health_code_update();
+
/* Lock stream because we are about to change its state. */
pthread_mutex_lock(&stream->lock);
stream->net_seq_idx = relayd_id;
DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
stream->name, stream->key);
}
+ if (relayd_id != -1ULL) {
+ ret = consumer_send_relayd_streams_sent(relayd_id);
+ if (ret < 0) {
+ goto error_unlock;
+ }
+ }
ustctl_flush_buffer(stream->ustream, 1);
ssize_t read_len;
unsigned long len, padded_len;
+ health_code_update();
+
DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
uint64_t len, struct lttng_consumer_channel *channel,
int timer, int wait)
{
- int ret, ret_code = LTTNG_OK;
+ int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
char *metadata_str;
DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
goto end;
}
+ health_code_update();
+
/* Receive metadata string. */
ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
if (ret < 0) {
goto end_free;
}
+ health_code_update();
+
pthread_mutex_lock(&channel->metadata_cache->lock);
ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
if (ret < 0) {
}
while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
DBG("Waiting for metadata to be flushed");
+
+ health_code_update();
+
usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
}
int sock, struct pollfd *consumer_sockpoll)
{
ssize_t ret;
- enum lttng_error_code ret_code = LTTNG_OK;
+ enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
struct lttcomm_consumer_msg msg;
struct lttng_consumer_channel *channel = NULL;
+ health_code_update();
+
ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
if (ret != sizeof(msg)) {
DBG("Consumer received unexpected message size %zd (expects %zu)",
}
return ret;
}
+
+ health_code_update();
+
if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
/*
* Notify the session daemon that the command is completed.
return -ENOENT;
}
+ health_code_update();
+
/* relayd needs RCU read-side lock */
rcu_read_lock();
attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
attr.chan_id = msg.u.ask_channel.chan_id;
- attr.output = msg.u.ask_channel.output;
memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
+ /* Match channel buffer type to the UST abi. */
+ switch (msg.u.ask_channel.output) {
+ case LTTNG_EVENT_MMAP:
+ default:
+ attr.output = LTTNG_UST_MMAP;
+ break;
+ }
+
/* Translate and save channel type. */
switch (msg.u.ask_channel.type) {
case LTTNG_UST_CHAN_PER_CPU:
goto error_fatal;
};
+ health_code_update();
+
ret = ask_channel(ctx, sock, channel, &attr);
if (ret < 0) {
goto end_channel_error;
msg.u.ask_channel.live_timer_interval);
}
+ health_code_update();
+
/*
* Add the channel to the internal state AFTER all streams were created
* and successfully sent to session daemon. This way, all streams must
goto end_channel_error;
}
+ health_code_update();
+
/*
* Channel and streams are now created. Inform the session daemon that
* everything went well and should wait to receive the channel and
goto end_msg_sessiond;
}
+ health_code_update();
+
/* Send everything to sessiond. */
ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
if (ret < 0) {
goto error_fatal;
}
+ health_code_update();
+
/*
* In no monitor mode, the streams ownership is kept inside the channel
* so don't send them to the data thread.
channel = consumer_find_channel(key);
if (!channel) {
- ERR("UST consumer push metadata %" PRIu64 " not found", key);
- ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
+ /*
+ * This is possible if the metadata creation on the consumer side
+ * is in flight vis-a-vis a concurrent push metadata from the
+ * session daemon. Simply return that the channel failed and the
+ * session daemon will handle that message correctly considering
+ * that this race is acceptable thus the DBG() statement here.
+ */
+ DBG("UST consumer push metadata %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
goto end_msg_sessiond;
}
+ health_code_update();
+
/* Tell session daemon we are ready to receive the metadata. */
- ret = consumer_send_status_msg(sock, LTTNG_OK);
+ ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
if (ret < 0) {
/* Somehow, the session daemon is not responding anymore. */
goto error_fatal;
}
+ health_code_update();
+
/* Wait for more data. */
- if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
+ health_poll_entry();
+ ret = lttng_consumer_poll_socket(consumer_sockpoll);
+ health_poll_exit();
+ if (ret < 0) {
goto error_fatal;
}
+ health_code_update();
+
ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
len, channel, 0, 1);
if (ret < 0) {
}
}
+ health_code_update();
ret = consumer_send_status_msg(sock, ret_code);
if (ret < 0) {
/* Somehow, the session daemon is not responding anymore. */
goto end_nosignal;
}
+ health_code_update();
break;
}
default:
end_nosignal:
rcu_read_unlock();
+ health_code_update();
+
/*
* Return 1 to indicate success since the 0 value can be a socket
* shutdown during the recv() or send() call.
goto error_fatal;
}
rcu_read_unlock();
+
+ health_code_update();
+
return 1;
end_channel_error:
if (channel) {
goto error_fatal;
}
rcu_read_unlock();
+
+ health_code_update();
+
return 1;
error_fatal:
rcu_read_unlock();
*
* Return 0 on success or else a negative value.
*/
-static int get_index_values(struct lttng_packet_index *index,
+static int get_index_values(struct ctf_packet_index *index,
struct ustctl_consumer_stream *ustream)
{
int ret;
long ret = 0;
char dummy;
struct ustctl_consumer_stream *ustream;
- struct lttng_packet_index index;
+ struct ctf_packet_index index;
assert(stream);
assert(stream->ustream);
/* Ease our life for what's next. */
ustream = stream->ustream;
- /* We can consume the 1 byte written into the wait_fd by UST */
+ /*
+ * We can consume the 1 byte written into the wait_fd by UST.
+ * Don't trigger error if we cannot read this one byte (read
+ * returns 0), or if the error is EAGAIN or EWOULDBLOCK.
+ */
if (stream->monitor && !stream->hangup_flush_done) {
ssize_t readlen;
- do {
- readlen = read(stream->wait_fd, &dummy, 1);
- } while (readlen == -1 && errno == EINTR);
- if (readlen == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
+ readlen = lttng_read(stream->wait_fd, &dummy, 1);
+ if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
ret = readlen;
goto end;
}
rcu_read_lock();
cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
node.node) {
+
+ health_code_update();
+
pthread_mutex_lock(&stream->chan->lock);
/*
* Whatever returned value, we must continue to try to close everything
{
struct lttcomm_metadata_request_msg request;
struct lttcomm_consumer_msg msg;
- enum lttng_error_code ret_code = LTTNG_OK;
+ enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
uint64_t len, key, offset;
int ret;
request.key);
pthread_mutex_lock(&ctx->metadata_socket_lock);
+
+ health_code_update();
+
ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
sizeof(request));
if (ret < 0) {
goto end;
}
+ health_code_update();
+
/* Receive the metadata from sessiond */
ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
sizeof(msg));
goto end;
}
+ health_code_update();
+
if (msg.cmd_type == LTTNG_ERR_UND) {
/* No registry found */
(void) consumer_send_status_msg(ctx->consumer_metadata_socket,
DBG("No new metadata to receive for key %" PRIu64, key);
}
+ health_code_update();
+
/* Tell session daemon we are ready to receive the metadata. */
ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
- LTTNG_OK);
+ LTTCOMM_CONSUMERD_SUCCESS);
if (ret < 0 || len == 0) {
/*
* Somehow, the session daemon is not responding anymore or there is
goto end;
}
- ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
+ health_code_update();
+
+ ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
key, offset, len, channel, timer, wait);
- if (ret_code >= 0) {
+ if (ret >= 0) {
/*
* Only send the status msg if the sessiond is alive meaning a positive
* ret code.
*/
- (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
+ (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
}
ret = 0;
end:
+ health_code_update();
+
pthread_mutex_unlock(&ctx->metadata_socket_lock);
return ret;
}
projects / lttng-tools.git / blobdiff