// There is currently the possibility of a race condition if a chunk
// upload timed-out. It's possible that a second upload succeeds,
// has the chunk marked as "complete" and then the first request makes
// its way through the queue and marks the chunk as pending again.
// Since we are just about to close the file, we'll check to see if any
// chunks are marked as pending, and if so, we'll retry them.
void check_for_complete_chunks(vector<File> &files) {
for (int currCheckNum=0; currCheckNum < NUM_CHUNK_CHECKS; ++currCheckNum){
map<string, JSON> fileDescriptions;
while (!chunksFinished.empty()) {
Chunk *c = chunksFinished.consume();
// Cache file descriptions so we only have to do once per file,
// not once per chunk.
if (fileDescriptions.find(c->fileID) == fileDescriptions.end())
fileDescriptions[c->fileID] = fileDescribe(c->fileID);
if (!is_chunk_complete(c, fileDescriptions[c->fileID])) {
// After the chunk was uploaded, it was cleared, removing the data
// from the buffer. We need to reload if we're going to upload again.
chunksToRead.produce(c);
}
}
// All of the chunks were marked as complete, so let's exit and we
// should be safeish to close the file.
if(chunksToRead.size() == 0)
return;
// Set the totalChunks variable to the # of chunks we're going
// to retry now plus the number of chunks in the failed queue. The monitor
// thread will be busy until the size of chunksFinished + chunksFailed
// equals totalChunks.
DXLOG(logINFO) << "Retrying " << chunksToRead.size() << " chunks that did not complete.";
totalChunks = chunksToRead.size() + chunksFailed.size();
// Read, compress, and upload the chunks which weren't marked as complete.
createWorkerThreads(files);
boost::thread monitorThread(monitor);
monitorThread.join();
interruptWorkerThreads();
joinWorkerThreads();
}
// We have tried to upload incomplete chunks NUM_CHUNK_CHECKS times!
// Check to see if there are any chunks still not complete and if so,
// print warning.
map<string, JSON> fileDescriptions;
while (!chunksFinished.empty()) {
Chunk *c = chunksFinished.consume();
// Cache file descriptions so we only have to do once per file,
// not once per chunk.
if (fileDescriptions.find(c->fileID) == fileDescriptions.end())
fileDescriptions[c->fileID] = fileDescribe(c->fileID);
if (!is_chunk_complete(c, fileDescriptions[c->fileID])) {
cerr << "Chunk " << c->index << " of file " << c->fileID << " did not complete. This file will not be accessible. PLease try to upload this file again." << endl;
}
}
}
void compressChunks() {
try {
while (true) {
Chunk * c = chunksToCompress.consume();
if (c->toCompress) {
c->log("Compressing...");
c->compress();
c->log("Finished compressing");
} else {
c->log("Not compressing");
}
chunksToUpload.produce(c);
// Sleep for tiny amount of time, to make sure we yield to other threads.
// Note: boost::this_thread::yield() is not a valid interruption point,
// so we have to use sleep()
boost::this_thread::sleep(boost::posix_time::microseconds(100));
}
} catch(std::bad_alloc &e) {
boost::call_once(bad_alloc_once, boost::bind(&handle_bad_alloc, e));
} catch (boost::thread_interrupted &ti) {
return;
}
}
void verifyChunkMD5(vector<File> &files) {
try {
while (true) {
Chunk * c = chunksToComputeMD5.consume();
if (files[c->parentFileIndex].matchStatus == File::Status::FAILED_TO_MATCH_REMOTE_FILE) {
// We have already marked file as a non-match, don't waste time reading more chunks from it
c->log("File status == FAILED_TO_MATCH_REMOTE_FILE, Skipping the MD5 compute...");
c->clear();
chunksSkipped.produce(c);
} else {
c->log("Computing MD5...");
string computedMD5 = c->computeMD5();
c->clear();
if (c->expectedMD5 != computedMD5) {
c->log("MISMATCH between expected MD5 '" + c->expectedMD5 + "', and computed MD5 '" + computedMD5 + "' ... marking the file as Mismatch");
files[c->parentFileIndex].matchStatus = File::Status::FAILED_TO_MATCH_REMOTE_FILE;
chunksFailed.produce(c);
} else {
c->log("Expected and computed MD5 match!");
chunksFinished.produce(c);
}
}
}
} catch (boost::thread_interrupted &ti) {
return;
}
}
void readChunks(const vector<File> &files) {
try {
while (true) {
Chunk * c = chunksToRead.consume();
if (files[c->parentFileIndex].matchStatus == File::Status::FAILED_TO_MATCH_REMOTE_FILE) {
// We have already marked file as a non-match, don't waste time reading more chunks from it
c->log("File status == FAILED_TO_MATCH_REMOTE_FILE, Skipping the read...");
chunksSkipped.produce(c);
} else {
c->log("Reading...");
c->read();
c->log("Finished reading");
chunksToComputeMD5.produce(c);
}
}
} catch (boost::thread_interrupted &ti) {
return;
}
}
int main(int argc, char * argv[]) {
try {
// Note: Verbose mode logging is enabled (if requested) by options parse()
opt.parse(argc, argv);
} catch (exception &e) {
cerr << "Error processing arguments: " << e.what() << endl;
opt.printHelp(argv[0]);
return 1;
}
if (opt.env()) {
opt.setApiserverDxConfig(); // needed for 'ua --env' to report project name
printEnvironmentInfo();
return 0;
}
if (opt.version()) {
cout << "Upload Agent Version: " << UAVERSION;
#if OLD_KERNEL_SUPPORT
cout << " (old-kernel-support)";
#endif
cout << endl
<< "git version: " << DXTOOLKIT_GITVERSION << endl
<< "libboost version: " << (BOOST_VERSION / 100000) << "." << ((BOOST_VERSION / 100) % 1000) << "." << (BOOST_VERSION % 100) << endl
<< "libcurl version: " << LIBCURL_VERSION_MAJOR << "." << LIBCURL_VERSION_MINOR << "." << LIBCURL_VERSION_PATCH << endl;
return 0;
} else if (opt.help() || opt.files.empty()) {
opt.printHelp(argv[0]);
return (opt.help()) ? 0 : 1;
}
setUserAgentString(); // also sets dx::config::USER_AGENT_STRING()
DXLOG(logINFO) << "DNAnexus Upload Agent " << UAVERSION << " (git version: " << DXTOOLKIT_GITVERSION << ")";
DXLOG(logINFO) << "Upload agent's User Agent string: '" << userAgentString << "'";
DXLOG(logINFO) << "dxcpp's User Agent string: '" << dx::config::USER_AGENT_STRING() << "'";
DXLOG(logINFO) << opt;
try {
opt.setApiserverDxConfig();
opt.validate();
/*
* Check for updates, and terminate execution if necessary. This also
* has the side effect of verifying that we can connect to the API
* server, and that the authentication token is valid.
*/
try {
checkForUpdates();
} catch (runtime_error &e) {
cerr << endl << e.what() << endl;
return 3;
}
if (!opt.doNotResume) {
disallowDuplicateFiles(opt.files, opt.projects);
}
} catch (exception &e) {
cerr << endl << "ERROR: " << e.what() << endl;
return 1;
}
const bool anyImportAppToBeCalled = (opt.reads || opt.pairedReads || opt.mappings || opt.variants);
chunksToCompress.setCapacity(opt.compressThreads);
chunksToUpload.setCapacity(opt.uploadThreads);
int exitCode = 0;
try {
curlInit(); // for curl requests to be made by upload chunk request
NUMTRIES_g = opt.tries;
vector<File> files;
for (unsigned int i = 0; i < opt.files.size(); ++i) {
DXLOG(logINFO) << "Getting MIME type for local file " << opt.files[i] << "...";
string mimeType = getMimeType(opt.files[i]);
DXLOG(logINFO) << "MIME type for local file " << opt.files[i] << " is '" << mimeType << "'.";
bool toCompress;
if (!opt.doNotCompress) {
bool is_compressed = isCompressed(mimeType);
toCompress = !is_compressed;
if (is_compressed)
DXLOG(logINFO) << "File " << opt.files[i] << " is already compressed, so won't try to compress it any further.";
else
DXLOG(logINFO) << "File " << opt.files[i] << " is not compressed, will compress it before uploading.";
} else {
toCompress = false;
}
if (toCompress) {
mimeType = "application/x-gzip";
}
files.push_back(File(opt.files[i], opt.projects[i], opt.folders[i], opt.names[i], toCompress, !opt.doNotResume, mimeType, opt.chunkSize, i));
totalChunks += files[i].createChunks(chunksToRead, opt.tries);
cerr << endl;
}
if (opt.waitOnClose) {
for (unsigned int i = 0; i < files.size(); ++i) {
files[i].waitOnClose = true;
}
}
// Create folders all at once (instead of one by one, above, where we
// initialize the File objects).
createFolders(opt.projects, opt.folders);
// Take this point as the starting time for program operation
// (to calculate average transfer speed)
startTime = std::time(0);
DXLOG(logINFO) << "Created " << totalChunks << " chunks.";
createWorkerThreads(files);
DXLOG(logINFO) << "Creating monitor thread..";
boost::thread monitorThread(monitor);
boost::thread uploadProgressThread;
if (opt.progress) {
DXLOG(logINFO) << "Creating Upload Progress thread..";
uploadProgressThread = boost::thread(uploadProgress, boost::ref(files));
}
DXLOG(logINFO) << "Joining monitor thread...";
monitorThread.join();
DXLOG(logINFO) << "Monitor thread finished.";
if (opt.progress) {
DXLOG(logINFO) << "Joining Upload Progress thread..";
keepShowingUploadProgress = false;
uploadProgressThread.interrupt();
uploadProgressThread.join();
DXLOG(logINFO) << "Upload Progress thread finished.";
}
interruptWorkerThreads();
joinWorkerThreads();
while (!chunksFailed.empty()) {
Chunk * c = chunksFailed.consume();
c->log("Chunk failed", logERROR);
markFileAsFailed(files, c->fileID);
}
if (opt.verbose) {
cerr << endl;
}
for (unsigned int i = 0; i < files.size(); ++i) {
if (files[i].failed) {
cerr << "File \""<< files[i].localFile << "\" could not be uploaded." << endl;
} else {
cerr << "File \"" << files[i].localFile << "\" was uploaded successfully. Closing..." << endl;
if (files[i].isRemoteFileOpen) {
files[i].close();
}
}
if (files[i].failed)
files[i].fileID = "failed";
}
DXLOG(logINFO) << "Waiting for files to be closed...";
boost::thread waitOnCloseThread(waitOnClose, boost::ref(files));
DXLOG(logINFO) << "Joining wait-on-close thread...";
waitOnCloseThread.join();
DXLOG(logINFO) << "Wait-on-close thread finished.";
if (anyImportAppToBeCalled) {
runImportApps(opt, files);
}
for (unsigned i = 0; i < files.size(); ++i) {
cout << files[i].fileID;
if (files[i].fileID == "failed")
exitCode = 1;
if (anyImportAppToBeCalled) {
if (files[i].jobID == "failed")
exitCode = 1;
cout << "\t" << files[i].jobID;
}
cout << endl;
}
curlCleanup();
DXLOG(logINFO) << "Exiting.";
} catch (bad_alloc &e) {
boost::call_once(bad_alloc_once, boost::bind(&handle_bad_alloc, e));
} catch (exception &e) {
curlCleanup();
cerr << endl << "ERROR: " << e.what() << endl;
return 1;
}
return exitCode;
}
void uploadChunks(vector<File> &files) {
try {
while (true) {
Chunk * c = chunksToUpload.consume();
c->log("Uploading...");
bool uploaded = false;
try {
c->upload(opt);
uploaded = true;
} catch (runtime_error &e) {
ostringstream msg;
msg << "Upload failed: " << e.what();
c->log(msg.str(), logERROR);
}
if (uploaded) {
c->log("Upload succeeded!");
int64_t size_of_chunk = c->data.size(); // this can be different than (c->end - c->start) because of compression
c->clear();
chunksFinished.produce(c);
// Update number of bytes uploaded in parent file object
boost::mutex::scoped_lock boLock(bytesUploadedMutex);
files[c->parentFileIndex].bytesUploaded += (c->end - c->start);
files[c->parentFileIndex].atleastOnePartDone = true;
bytesUploadedSinceStart += size_of_chunk;
boLock.unlock();
} else if (c->triesLeft > 0) {
int numTry = NUMTRIES_g - c->triesLeft + 1; // find out which try is it
int timeout = (numTry > 6) ? 256 : 4 << numTry; // timeout is always between [8, 256] seconds
c->log("Will retry reading and uploading this chunks in " + boost::lexical_cast<string>(timeout) + " seconds", logWARNING);
if (!opt.noRoundRobinDNS) {
boost::mutex::scoped_lock forceRefreshLock(forceRefreshDNSMutex);
c->log("Setting forceRefreshDNS = true in main.cpp:uploadChunks()");
forceRefreshDNS = true; // refresh the DNS list in next call to getRandomIP()
}
--(c->triesLeft);
c->clear(); // we will read & compress data again
boost::this_thread::sleep(boost::posix_time::milliseconds(timeout * 1000));
// We push the chunk to retry to "chunksToRead" and not "chunksToUpload"
// Since chunksToUpload queue is bounded, and chunksToUpload.produce() can block,
// thus giving rise to deadlock
chunksToRead.produce(c);
} else {
c->log("Not retrying", logERROR);
// TODO: Should we print it on stderr or DXLOG (verbose only) ??
cerr << "\nFailed to upload Chunk [" << c->start << " - " << c->end << "] for local file ("
<< files[c->parentFileIndex].localFile << "). APIServer response for last try: '" << c->respData << "'" << endl;
c->clear();
chunksFailed.produce(c);
}
// Sleep for tiny amount of time, to make sure we yield to other threads.
// Note: boost::this_thread::yield() is not a valid interruption point,
// so we have to use sleep()
boost::this_thread::sleep(boost::posix_time::microseconds(100));
}
} catch(std::bad_alloc &e) {
boost::call_once(bad_alloc_once, boost::bind(&handle_bad_alloc, e));
} catch (boost::thread_interrupted &ti) {
return;
}
}
