X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=httpinput.cpp;h=bf8a8b1dd22e609d6611193ae1cf327c96c3d758;hb=37bc14f5b28f66ccc1760f410091f98769616476;hp=80daa1a8a220e1b5b4d93ca3be99145dc2f35854;hpb=979a284b4039b0ea74525b700b9f1089b8c4248d;p=cubemap diff --git a/httpinput.cpp b/httpinput.cpp index 80daa1a..bf8a8b1 100644 --- a/httpinput.cpp +++ b/httpinput.cpp @@ -8,6 +8,7 @@ #include #include #include +#include #include #include #include @@ -22,47 +23,53 @@ #include "parse.h" #include "serverpool.h" #include "state.pb.h" +#include "stream.h" +#include "timespec.h" #include "util.h" #include "version.h" using namespace std; -extern ServerPool *servers; - namespace { -// Compute b-a. -timespec clock_diff(const timespec &a, const timespec &b) +string host_header(const string &host, const string &port) { - 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; + if (port == "http" || atoi(port.c_str()) == 80) { + return host; + } else { + return host + ":" + port; } - assert(ret.tv_nsec >= 0); - return ret; } } // namespace -HTTPInput::HTTPInput(const string &url) +extern ServerPool *servers; + +HTTPInput::HTTPInput(const string &url, Input::Encoding encoding) : state(NOT_CONNECTED), url(url), + encoding(encoding), has_metacube_header(false), - sock(-1) + sock(-1), + num_connection_attempts(0), + suppress_logging(false) { pthread_mutex_init(&stats_mutex, NULL); stats.url = url; stats.bytes_received = 0; stats.data_bytes_received = 0; stats.connect_time = -1; + + last_verbose_connection.tv_sec = -3600; + last_verbose_connection.tv_nsec = 0; } HTTPInput::HTTPInput(const InputProto &serialized) : state(State(serialized.state())), url(serialized.url()), + encoding(serialized.is_metacube_encoded() ? + Input::INPUT_ENCODING_METACUBE : + Input::INPUT_ENCODING_RAW), request(serialized.request()), request_bytes_sent(serialized.request_bytes_sent()), response(serialized.response()), @@ -74,15 +81,8 @@ HTTPInput::HTTPInput(const InputProto &serialized) pending_data.resize(serialized.pending_data().size()); memcpy(&pending_data[0], serialized.pending_data().data(), serialized.pending_data().size()); - string protocol; - parse_url(url, &protocol, &host, &port, &path); // Don't care if it fails. - - // Older versions stored the extra \r\n in the HTTP header. - // Strip it if we find it. - if (http_header.size() >= 4 && - memcmp(http_header.data() + http_header.size() - 4, "\r\n\r\n", 4) == 0) { - http_header.resize(http_header.size() - 2); - } + string protocol, user; + parse_url(url, &protocol, &user, &host, &port, &path); // Don't care if it fails. pthread_mutex_init(&stats_mutex, NULL); stats.url = url; @@ -99,6 +99,7 @@ void HTTPInput::close_socket() { if (sock != -1) { safe_close(sock); + sock = -1; } MutexLock lock(&stats_mutex); @@ -121,6 +122,12 @@ InputProto HTTPInput::serialize() const serialized.set_bytes_received(stats.bytes_received); serialized.set_data_bytes_received(stats.data_bytes_received); serialized.set_connect_time(stats.connect_time); + if (encoding == Input::INPUT_ENCODING_METACUBE) { + serialized.set_is_metacube_encoded(true); + } else { + assert(encoding == Input::INPUT_ENCODING_RAW); + serialized.set_is_metacube_encoded(false); + } return serialized; } @@ -204,53 +211,42 @@ int HTTPInput::lookup_and_connect(const string &host, const string &port) return -1; } -bool HTTPInput::parse_response(const std::string &request) +bool HTTPInput::parse_response(const string &request) { vector lines = split_lines(response); if (lines.empty()) { - log(WARNING, "[%s] Empty HTTP response from input.", url.c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Empty HTTP response from input.", url.c_str()); + } return false; } vector first_line_tokens = split_tokens(lines[0]); if (first_line_tokens.size() < 2) { - log(WARNING, "[%s] Malformed response line '%s' from input.", - url.c_str(), lines[0].c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Malformed response line '%s' from input.", + url.c_str(), lines[0].c_str()); + } return false; } int response = atoi(first_line_tokens[1].c_str()); if (response != 200) { - log(WARNING, "[%s] Non-200 response '%s' from input.", - url.c_str(), lines[0].c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Non-200 response '%s' from input.", + url.c_str(), lines[0].c_str()); + } return false; } - multimap parameters; - for (size_t i = 1; i < lines.size(); ++i) { - size_t split = lines[i].find(":"); - if (split == string::npos) { - log(WARNING, "[%s] Ignoring malformed HTTP response line '%s'", - url.c_str(), lines[i].c_str()); - continue; - } + multimap parameters = extract_headers(lines, url); - string key(lines[i].begin(), lines[i].begin() + split); - - // Skip any spaces after the colon. - do { - ++split; - } while (split < lines[i].size() && lines[i][split] == ' '); - - string value(lines[i].begin() + split, lines[i].end()); - - // Remove “Content-encoding: metacube”. - // TODO: Make case-insensitive. - if (key == "Content-encoding" && value == "metacube") { - continue; - } - - parameters.insert(make_pair(key, value)); + // Remove “Content-encoding: metacube”. + // TODO: Make case-insensitive. + multimap::iterator encoding_it = + parameters.find("Content-encoding"); + if (encoding_it != parameters.end() && encoding_it->second == "metacube") { + parameters.erase(encoding_it); } // Change “Server: foo” to “Server: metacube/0.1 (reflecting: foo)” @@ -295,7 +291,7 @@ void HTTPInput::do_work() // 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); + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); assert(err != -1); } @@ -305,13 +301,15 @@ void HTTPInput::do_work() static const int timeout_secs = 30; timespec now; - int err = clock_gettime(CLOCK_MONOTONIC, &now); + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &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); + if (!suppress_logging) { + log(ERROR, "[%s] Timeout after %d seconds, closing.", url.c_str(), elapsed.tv_sec); + } state = CLOSING_SOCKET; continue; } @@ -324,7 +322,7 @@ void HTTPInput::do_work() bool activity = wait_for_activity(sock, (state == SENDING_REQUEST) ? POLLOUT : POLLIN, &timeout); if (activity) { - err = clock_gettime(CLOCK_MONOTONIC, &last_activity); + err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); assert(err != -1); } else { // OK. Most likely, should_stop was set, or we have timed out. @@ -344,13 +342,40 @@ void HTTPInput::do_work() } { - string protocol; // Thrown away. - if (!parse_url(url, &protocol, &host, &port, &path)) { - log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str()); + string protocol, user; // Thrown away. + if (!parse_url(url, &protocol, &user, &host, &port, &path)) { + if (!suppress_logging) { + log(WARNING, "[%s] Failed to parse URL '%s'", url.c_str(), url.c_str()); + } break; } + + // Remove the brackets around IPv6 address literals. + // TODO: See if we can join this with the code in parse_ip_address(), + // or maybe even more it into parse_url(). + if (!host.empty() && host[0] == '[' && host[host.size() - 1] == ']') { + host = host.substr(1, host.size() - 2); + } } + if (suppress_logging) { + // See if there's more than one minute since last time we made a connection + // with logging enabled. If so, turn it on again. + timespec now; + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &now); + assert(err != -1); + + double elapsed = now.tv_sec - last_verbose_connection.tv_sec + + 1e-9 * (now.tv_nsec - last_verbose_connection.tv_nsec); + if (elapsed > 60.0) { + suppress_logging = false; + } + } + if (!suppress_logging) { + int err = clock_gettime(CLOCK_MONOTONIC_COARSE, &last_verbose_connection); + assert(err != -1); + } + ++num_connection_attempts; sock = lookup_and_connect(host, port); if (sock != -1) { // Yay, successful connect. Try to set it as nonblocking. @@ -360,13 +385,13 @@ void HTTPInput::do_work() state = CLOSING_SOCKET; } else { state = SENDING_REQUEST; - request = "GET " + path + " HTTP/1.0\r\nUser-Agent: cubemap\r\n\r\n"; + request = "GET " + path + " HTTP/1.0\r\nHost: " + host_header(host, port) + "\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); + clock_gettime(CLOCK_MONOTONIC_COARSE, &last_activity); } break; case SENDING_REQUEST: { @@ -407,8 +432,10 @@ void HTTPInput::do_work() if (ret == 0) { // This really shouldn't happen... - log(ERROR, "[%s] Socket unexpectedly closed while reading header", - url.c_str()); + if (!suppress_logging) { + log(ERROR, "[%s] Socket unexpectedly closed while reading header", + url.c_str()); + } state = CLOSING_SOCKET; continue; } @@ -416,7 +443,9 @@ void HTTPInput::do_work() RequestParseStatus status = wait_for_double_newline(&response, buf, ret); if (status == RP_OUT_OF_SPACE) { - log(WARNING, "[%s] Sever sent overlong HTTP response!", url.c_str()); + if (!suppress_logging) { + log(WARNING, "[%s] Server sent overlong HTTP response!", url.c_str()); + } state = CLOSING_SOCKET; continue; } else if (status == RP_NOT_FINISHED_YET) { @@ -443,8 +472,16 @@ void HTTPInput::do_work() process_data(&extra_data[0], extra_data.size()); } - log(INFO, "[%s] Connected to '%s', receiving data.", - url.c_str(), url.c_str()); + if (!suppress_logging) { + if (encoding == Input::INPUT_ENCODING_RAW) { + log(INFO, "[%s] Connected to '%s', receiving raw data.", + url.c_str(), url.c_str()); + } else { + assert(encoding == Input::INPUT_ENCODING_METACUBE); + log(INFO, "[%s] Connected to '%s', receiving data.", + url.c_str(), url.c_str()); + } + } state = RECEIVING_DATA; break; } @@ -464,12 +501,28 @@ void HTTPInput::do_work() if (ret == 0) { // This really shouldn't happen... - log(ERROR, "[%s] Socket unexpectedly closed while reading data", - url.c_str()); + if (!suppress_logging) { + log(ERROR, "[%s] Socket unexpectedly closed while reading data", + url.c_str()); + } state = CLOSING_SOCKET; continue; } + num_connection_attempts = 0; // Reset, since we have a successful read. + if (suppress_logging) { + // This was suppressed earlier, so print it out now. + if (encoding == Input::INPUT_ENCODING_RAW) { + log(INFO, "[%s] Connected to '%s', receiving raw data.", + url.c_str(), url.c_str()); + } else { + assert(encoding == Input::INPUT_ENCODING_METACUBE); + log(INFO, "[%s] Connected to '%s', receiving data.", + url.c_str(), url.c_str()); + } + suppress_logging = false; + } + process_data(buf, ret); break; } @@ -486,7 +539,15 @@ void HTTPInput::do_work() // or the connection just got closed. // The earlier steps have already given the error message, if any. if (state == NOT_CONNECTED && !should_stop()) { - log(INFO, "[%s] Waiting 0.2 second and restarting...", url.c_str()); + if (!suppress_logging) { + log(INFO, "[%s] Waiting 0.2 seconds and restarting...", url.c_str()); + } + + if (num_connection_attempts >= 3 && !suppress_logging) { + log(INFO, "[%s] %d failed connection attempts, suppressing logging for one minute.", + url.c_str(), num_connection_attempts); + suppress_logging = true; + } timespec timeout_ts; timeout_ts.tv_sec = 0; timeout_ts.tv_nsec = 200000000; @@ -497,12 +558,21 @@ void HTTPInput::do_work() 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; } + if (encoding == Input::INPUT_ENCODING_RAW) { + for (size_t i = 0; i < stream_indices.size(); ++i) { + servers->add_data(stream_indices[i], ptr, bytes, /*metacube_flags=*/0); + } + return; + } + + assert(encoding == Input::INPUT_ENCODING_METACUBE); + pending_data.insert(pending_data.end(), ptr, ptr + bytes); + for ( ;; ) { // If we don't have enough data (yet) for even the Metacube header, just return. if (pending_data.size() < sizeof(metacube2_block_header)) { @@ -535,16 +605,17 @@ void HTTPInput::process_data(char *ptr, size_t bytes) } // Now it's safe to read the header. - metacube2_block_header *hdr = reinterpret_cast(pending_data.data()); - assert(memcmp(hdr->sync, METACUBE2_SYNC, sizeof(hdr->sync)) == 0); - uint32_t size = ntohl(hdr->size); - uint16_t flags = ntohs(hdr->flags); - uint16_t expected_csum = metacube2_compute_crc(hdr); - - if (expected_csum != ntohs(hdr->csum)) { + metacube2_block_header hdr; + memcpy(&hdr, pending_data.data(), sizeof(hdr)); + assert(memcmp(hdr.sync, METACUBE2_SYNC, sizeof(hdr.sync)) == 0); + uint32_t size = ntohl(hdr.size); + uint16_t flags = ntohs(hdr.flags); + uint16_t expected_csum = metacube2_compute_crc(&hdr); + + if (expected_csum != ntohs(hdr.csum)) { log(WARNING, "[%s] Metacube checksum failed (expected 0x%x, got 0x%x), " "not reading block claiming to be %d bytes (flags=%x).", - url.c_str(), expected_csum, ntohs(hdr->csum), + url.c_str(), expected_csum, ntohs(hdr.csum), size, flags); // Drop only the first byte, and let the rest of the code handle resync. @@ -552,8 +623,8 @@ void HTTPInput::process_data(char *ptr, size_t bytes) has_metacube_header = false; continue; } - if (size > 262144) { - log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header?", + if (size > 10485760) { + log(WARNING, "[%s] Metacube block of %d bytes (flags=%x); corrupted header??", url.c_str(), size, flags); } @@ -573,16 +644,9 @@ void HTTPInput::process_data(char *ptr, size_t bytes) for (size_t i = 0; i < stream_indices.size(); ++i) { 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; - } - for (size_t i = 0; i < stream_indices.size(); ++i) { - servers->add_data(stream_indices[i], inner_data, size, suitable_for_stream_start); - } + } + for (size_t i = 0; i < stream_indices.size(); ++i) { + servers->add_data(stream_indices[i], inner_data, size, flags); } // Consume the block. This isn't the most efficient way of dealing with things @@ -598,7 +662,7 @@ void HTTPInput::drop_pending_data(size_t num_bytes) if (num_bytes == 0) { return; } - log(WARNING, "[%s] Dropping %lld junk bytes from stream, maybe it is not a Metacube2 stream?", + log(WARNING, "[%s] Dropping %lld junk bytes; not a Metacube2 stream, or data was dropped from the middle of the stream.", url.c_str(), (long long)num_bytes); assert(pending_data.size() >= num_bytes); pending_data.erase(pending_data.begin(), pending_data.begin() + num_bytes);