void ReplSetDistLockManager::unlock(const DistLockHandle& lockSessionID) {
auto unlockStatus = _catalog->unlock(lockSessionID);
if (!unlockStatus.isOK()) {
queueUnlock(lockSessionID);
} else {
LOG(0) << "distributed lock with " << LocksType::lockID() << ": " << lockSessionID
<< "' unlocked.";
}
}
void ReplSetDistLockManager::doTask() {
LOG(0) << "creating distributed lock ping thread for process " << _processID
<< " (sleeping for " << _pingInterval << ")";
Timer elapsedSincelastPing(_serviceContext->getTickSource());
Client::initThread("replSetDistLockPinger");
while (!isShutDown()) {
{
auto txn = cc().makeOperationContext();
auto pingStatus = _catalog->ping(txn.get(), _processID, Date_t::now());
if (!pingStatus.isOK()) {
warning() << "pinging failed for distributed lock pinger" << causedBy(pingStatus);
}
const milliseconds elapsed(elapsedSincelastPing.millis());
if (elapsed > 10 * _pingInterval) {
warning() << "Lock pinger for proc: " << _processID << " was inactive for "
<< elapsed << " ms";
}
elapsedSincelastPing.reset();
std::deque<DistLockHandle> toUnlockBatch;
{
stdx::unique_lock<stdx::mutex> lk(_mutex);
toUnlockBatch.swap(_unlockList);
}
for (const auto& toUnlock : toUnlockBatch) {
auto unlockStatus = _catalog->unlock(txn.get(), toUnlock);
if (!unlockStatus.isOK()) {
warning() << "Failed to unlock lock with " << LocksType::lockID() << ": "
<< toUnlock << causedBy(unlockStatus);
queueUnlock(toUnlock);
} else {
LOG(0) << "distributed lock with " << LocksType::lockID() << ": " << toUnlock
<< "' unlocked.";
}
if (isShutDown()) {
return;
}
}
}
stdx::unique_lock<stdx::mutex> lk(_mutex);
_shutDownCV.wait_for(lk, _pingInterval);
}
}
StatusWith<DistLockManager::ScopedDistLock> ReplSetDistLockManager::lock(
StringData name, StringData whyMessage, milliseconds waitFor, milliseconds lockTryInterval) {
Timer timer(_serviceContext->getTickSource());
Timer msgTimer(_serviceContext->getTickSource());
while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) {
OID lockSessionID = OID::gen();
string who = str::stream() << _processID << ":" << getThreadName();
auto lockExpiration = _lockExpiration;
MONGO_FAIL_POINT_BLOCK(setDistLockTimeout, customTimeout) {
const BSONObj& data = customTimeout.getData();
lockExpiration = stdx::chrono::milliseconds(data["timeoutMs"].numberInt());
}
LOG(1) << "trying to acquire new distributed lock for " << name
<< " ( lock timeout : " << durationCount<Milliseconds>(lockExpiration)
<< " ms, ping interval : " << durationCount<Milliseconds>(_pingInterval)
<< " ms, process : " << _processID << " )"
<< " with lockSessionID: " << lockSessionID << ", why: " << whyMessage;
auto lockResult =
_catalog->grabLock(name, lockSessionID, who, _processID, Date_t::now(), whyMessage);
auto status = lockResult.getStatus();
if (status.isOK()) {
// Lock is acquired since findAndModify was able to successfully modify
// the lock document.
LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID;
return ScopedDistLock(lockSessionID, this);
}
if (status != ErrorCodes::LockStateChangeFailed) {
// An error occurred but the write might have actually been applied on the
// other side. Schedule an unlock to clean it up just in case.
queueUnlock(lockSessionID);
return status;
}
// Get info from current lock and check if we can overtake it.
auto getLockStatusResult = _catalog->getLockByName(name);
const auto& getLockStatus = getLockStatusResult.getStatus();
if (!getLockStatusResult.isOK() && getLockStatus != ErrorCodes::LockNotFound) {
return getLockStatus;
}
// Note: Only attempt to overtake locks that actually exists. If lock was not
// found, use the normal grab lock path to acquire it.
if (getLockStatusResult.isOK()) {
auto currentLock = getLockStatusResult.getValue();
auto canOvertakeResult = canOvertakeLock(currentLock, lockExpiration);
if (!canOvertakeResult.isOK()) {
return canOvertakeResult.getStatus();
}
if (canOvertakeResult.getValue()) {
auto overtakeResult = _catalog->overtakeLock(name,
lockSessionID,
currentLock.getLockID(),
who,
_processID,
Date_t::now(),
whyMessage);
const auto& overtakeStatus = overtakeResult.getStatus();
if (overtakeResult.isOK()) {
// Lock is acquired since findAndModify was able to successfully modify
// the lock document.
LOG(0) << "lock '" << name << "' successfully forced";
LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID;
return ScopedDistLock(lockSessionID, this);
}
if (overtakeStatus != ErrorCodes::LockStateChangeFailed) {
// An error occurred but the write might have actually been applied on the
// other side. Schedule an unlock to clean it up just in case.
queueUnlock(lockSessionID);
return overtakeStatus;
}
}
}
LOG(1) << "distributed lock '" << name << "' was not acquired.";
if (waitFor == milliseconds::zero()) {
break;
}
// Periodically message for debugging reasons
if (msgTimer.seconds() > 10) {
LOG(0) << "waited " << timer.seconds() << "s for distributed lock " << name << " for "
<< whyMessage;
msgTimer.reset();
}
milliseconds timeRemaining =
std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis()));
sleepFor(std::min(lockTryInterval, timeRemaining));
}
return {ErrorCodes::LockBusy, str::stream() << "timed out waiting for " << name};
}
StatusWith<DistLockManager::ScopedDistLock> ReplSetDistLockManager::lock(
OperationContext* txn,
StringData name,
StringData whyMessage,
milliseconds waitFor,
milliseconds lockTryInterval) {
Timer timer(_serviceContext->getTickSource());
Timer msgTimer(_serviceContext->getTickSource());
// Counts how many attempts have been made to grab the lock, which have failed with network
// error. This value is reset for each lock acquisition attempt because these are
// independent write operations.
int networkErrorRetries = 0;
// Distributed lock acquisition works by tring to update the state of the lock to 'taken'. If
// the lock is currently taken, we will back off and try the acquisition again, repeating this
// until the lockTryInterval has been reached. If a network error occurs at each lock
// acquisition attempt, the lock acquisition will be retried immediately.
while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) {
const OID lockSessionID = OID::gen();
const string who = str::stream() << _processID << ":" << getThreadName();
auto lockExpiration = _lockExpiration;
MONGO_FAIL_POINT_BLOCK(setDistLockTimeout, customTimeout) {
const BSONObj& data = customTimeout.getData();
lockExpiration = stdx::chrono::milliseconds(data["timeoutMs"].numberInt());
}
LOG(1) << "trying to acquire new distributed lock for " << name
<< " ( lock timeout : " << durationCount<Milliseconds>(lockExpiration)
<< " ms, ping interval : " << durationCount<Milliseconds>(_pingInterval)
<< " ms, process : " << _processID << " )"
<< " with lockSessionID: " << lockSessionID << ", why: " << whyMessage;
auto lockResult = _catalog->grabLock(
txn, name, lockSessionID, who, _processID, Date_t::now(), whyMessage);
auto status = lockResult.getStatus();
if (status.isOK()) {
// Lock is acquired since findAndModify was able to successfully modify
// the lock document.
log() << "distributed lock '" << name << "' acquired for '" << whyMessage
<< "', ts : " << lockSessionID;
return ScopedDistLock(txn, lockSessionID, this);
}
// If a network error occurred, unlock the lock synchronously and try again
if (ShardRegistry::kAllRetriableErrors.count(status.code()) &&
networkErrorRetries < kMaxNumLockAcquireRetries) {
LOG(1) << "Failed to acquire distributed lock because of retriable error. Retrying "
"acquisition by first unlocking the stale entry, which possibly exists now"
<< causedBy(status);
networkErrorRetries++;
status = _catalog->unlock(txn, lockSessionID);
if (status.isOK()) {
// We certainly do not own the lock, so we can retry
continue;
}
// Fall-through to the error checking logic below
invariant(status != ErrorCodes::LockStateChangeFailed);
LOG(1)
<< "Failed to retry acqusition of distributed lock. No more attempts will be made"
<< causedBy(status);
}
if (status != ErrorCodes::LockStateChangeFailed) {
// An error occurred but the write might have actually been applied on the
// other side. Schedule an unlock to clean it up just in case.
queueUnlock(lockSessionID);
return status;
}
// Get info from current lock and check if we can overtake it.
auto getLockStatusResult = _catalog->getLockByName(txn, name);
const auto& getLockStatus = getLockStatusResult.getStatus();
if (!getLockStatusResult.isOK() && getLockStatus != ErrorCodes::LockNotFound) {
return getLockStatus;
}
// Note: Only attempt to overtake locks that actually exists. If lock was not
// found, use the normal grab lock path to acquire it.
if (getLockStatusResult.isOK()) {
auto currentLock = getLockStatusResult.getValue();
auto canOvertakeResult = canOvertakeLock(txn, currentLock, lockExpiration);
if (!canOvertakeResult.isOK()) {
return canOvertakeResult.getStatus();
}
if (canOvertakeResult.getValue()) {
auto overtakeResult = _catalog->overtakeLock(txn,
name,
lockSessionID,
currentLock.getLockID(),
who,
_processID,
Date_t::now(),
whyMessage);
const auto& overtakeStatus = overtakeResult.getStatus();
if (overtakeResult.isOK()) {
// Lock is acquired since findAndModify was able to successfully modify
// the lock document.
LOG(0) << "lock '" << name << "' successfully forced";
LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID;
return ScopedDistLock(txn, lockSessionID, this);
}
if (overtakeStatus != ErrorCodes::LockStateChangeFailed) {
// An error occurred but the write might have actually been applied on the
// other side. Schedule an unlock to clean it up just in case.
queueUnlock(lockSessionID);
return overtakeStatus;
}
}
}
LOG(1) << "distributed lock '" << name << "' was not acquired.";
if (waitFor == milliseconds::zero()) {
break;
}
// Periodically message for debugging reasons
if (msgTimer.seconds() > 10) {
LOG(0) << "waited " << timer.seconds() << "s for distributed lock " << name << " for "
<< whyMessage;
msgTimer.reset();
}
// A new lock acquisition attempt will begin now (because the previous found the lock to be
// busy, so reset the retries counter)
networkErrorRetries = 0;
const milliseconds timeRemaining =
std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis()));
sleepFor(std::min(lockTryInterval, timeRemaining));
}
return {ErrorCodes::LockBusy, str::stream() << "timed out waiting for " << name};
}
