#include "httpinput.h"
#include "log.h"
#include "metacube.h"
+#include "mutexlock.h"
#include "parse.h"
#include "serverpool.h"
#include "state.pb.h"
using namespace std;
extern ServerPool *servers;
-
+
+namespace {
+
+// Compute b-a.
+timespec clock_diff(const timespec &a, const timespec &b)
+{
+ timespec ret;
+ ret.tv_sec = b.tv_sec - a.tv_sec;
+ ret.tv_nsec = b.tv_nsec - a.tv_nsec;
+ if (ret.tv_nsec < 0) {
+ ret.tv_sec--;
+ ret.tv_nsec += 1000000000;
+ }
+ assert(ret.tv_nsec >= 0);
+ return ret;
+}
+
+} // namespace
+
HTTPInput::HTTPInput(const string &url)
: state(NOT_CONNECTED),
url(url),
has_metacube_header(false),
sock(-1)
{
+ pthread_mutex_init(&stats_mutex, NULL);
+ stats.url = url;
+ stats.bytes_received = 0;
+ stats.data_bytes_received = 0;
+ stats.connect_time = -1;
}
HTTPInput::HTTPInput(const InputProto &serialized)
request_bytes_sent(serialized.request_bytes_sent()),
response(serialized.response()),
http_header(serialized.http_header()),
+ stream_header(serialized.stream_header()),
has_metacube_header(serialized.has_metacube_header()),
sock(serialized.sock())
{
memcmp(http_header.data() + http_header.size() - 4, "\r\n\r\n", 4) == 0) {
http_header.resize(http_header.size() - 2);
}
+
+ pthread_mutex_init(&stats_mutex, NULL);
+ stats.url = url;
+ stats.bytes_received = serialized.bytes_received();
+ stats.data_bytes_received = serialized.data_bytes_received();
+ if (serialized.has_connect_time()) {
+ stats.connect_time = serialized.connect_time();
+ } else {
+ stats.connect_time = time(NULL);
+ }
}
void HTTPInput::close_socket()
if (sock != -1) {
safe_close(sock);
}
+
+ MutexLock lock(&stats_mutex);
+ stats.connect_time = -1;
}
InputProto HTTPInput::serialize() const
serialized.set_request_bytes_sent(request_bytes_sent);
serialized.set_response(response);
serialized.set_http_header(http_header);
+ serialized.set_stream_header(stream_header);
serialized.set_pending_data(string(pending_data.begin(), pending_data.end()));
serialized.set_has_metacube_header(has_metacube_header);
serialized.set_sock(sock);
+ serialized.set_bytes_received(stats.bytes_received);
+ serialized.set_data_bytes_received(stats.data_bytes_received);
+ serialized.set_connect_time(stats.connect_time);
return serialized;
}
}
for (size_t i = 0; i < stream_indices.size(); ++i) {
- servers->set_header(stream_indices[i], http_header, "");
+ servers->set_header(stream_indices[i], http_header, stream_header);
}
return true;
void HTTPInput::do_work()
{
+ timespec last_activity;
+
+ // TODO: Make the timeout persist across restarts.
+ if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
+ int err = clock_gettime(CLOCK_MONOTONIC, &last_activity);
+ assert(err != -1);
+ }
+
while (!should_stop()) {
if (state == SENDING_REQUEST || state == RECEIVING_HEADER || state == RECEIVING_DATA) {
- bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, NULL);
- if (!activity) {
- // Most likely, should_stop was set.
+ // Give the socket 30 seconds since last activity before we time out.
+ static const int timeout_secs = 30;
+
+ timespec now;
+ int err = clock_gettime(CLOCK_MONOTONIC, &now);
+ assert(err != -1);
+
+ timespec elapsed = clock_diff(last_activity, now);
+ if (elapsed.tv_sec >= timeout_secs) {
+ // Timeout!
+ log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec);
+ state = CLOSING_SOCKET;
+ continue;
+ }
+
+ // Basically calculate (30 - (now - last_activity)) = (30 + (last_activity - now)).
+ // Add a second of slack to account for differences between clocks.
+ timespec timeout = clock_diff(now, last_activity);
+ timeout.tv_sec += timeout_secs + 1;
+ assert(timeout.tv_sec > 0 || (timeout.tv_sec >= 0 && timeout.tv_nsec > 0));
+
+ bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout);
+ if (activity) {
+ err = clock_gettime(CLOCK_MONOTONIC, &last_activity);
+ assert(err != -1);
+ } else {
+ // OK. Most likely, should_stop was set, or we have timed out.
continue;
}
}
request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n";
request_bytes_sent = 0;
}
+
+ MutexLock lock(&stats_mutex);
+ stats.connect_time = time(NULL);
+ clock_gettime(CLOCK_MONOTONIC, &last_activity);
}
break;
case SENDING_REQUEST: {
void HTTPInput::process_data(char *ptr, size_t bytes)
{
pending_data.insert(pending_data.end(), ptr, ptr + bytes);
+ {
+ MutexLock mutex(&stats_mutex);
+ stats.bytes_received += bytes;
+ }
for ( ;; ) {
// If we don't have enough data (yet) for even the Metacube header, just return.
uint32_t size = ntohl(hdr->size);
uint32_t flags = ntohl(hdr->flags);
- if (size > 65535) {
- log(WARNING, "[%s] Metacube block of %x bytes (flags=%x); corrupted header?",
+ if (size > 262144) {
+ log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?",
url.c_str(), size, flags);
}
return;
}
- // Send this block on to the data.
+ // Send this block on to the servers.
+ {
+ MutexLock lock(&stats_mutex);
+ stats.data_bytes_received += size;
+ }
char *inner_data = pending_data.data() + sizeof(metacube_block_header);
if (flags & METACUBE_FLAGS_HEADER) {
- string header(inner_data, inner_data + size);
+ stream_header = string(inner_data, inner_data + size);
for (size_t i = 0; i < stream_indices.size(); ++i) {
- servers->set_header(stream_indices[i], http_header, header);
+ servers->set_header(stream_indices[i], http_header, stream_header);
+ }
+ } else {
+ StreamStartSuitability suitable_for_stream_start;
+ if (flags & METACUBE_FLAGS_NOT_SUITABLE_FOR_STREAM_START) {
+ suitable_for_stream_start = NOT_SUITABLE_FOR_STREAM_START;
+ } else {
+ suitable_for_stream_start = SUITABLE_FOR_STREAM_START;
}
- } else {
for (size_t i = 0; i < stream_indices.size(); ++i) {
- servers->add_data(stream_indices[i], inner_data, size);
+ servers->add_data(stream_indices[i], inner_data, size, suitable_for_stream_start);
}
}
pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);
}
+void HTTPInput::add_destination(int stream_index)
+{
+ stream_indices.push_back(stream_index);
+ servers->set_header(stream_index, http_header, stream_header);
+}
+
+InputStats HTTPInput::get_stats() const
+{
+ MutexLock lock(&stats_mutex);
+ return stats;
+}