int move_range(const std::string &min_key, const std::string &max_key, int limit, std::string *moved_max_key){
// get key range
std::vector<std::string> keys;
ssdb::Status s;
s = src->keys(min_key, max_key, limit, &keys);
if(!s.ok()){
log_error("response error: %s", s.code().c_str());
return -1;
}
if(keys.empty()){
return 0;
}
if(moved_max_key){
*moved_max_key = keys[keys.size() - 1];
// lock key range
log_info("lock range %s", KeyRange(min_key, *moved_max_key).str().c_str());
const std::vector<std::string>* resp;
resp = src->request("set_kv_range", *moved_max_key, max_key);
if(!resp || resp->empty() || resp->at(0) != "ok"){
log_error("src server set_kv_range error!");
return -1;
}
}
// move key range
for(int i=0; i<(int)keys.size(); i++){
const std::string &key = keys[i];
if(move_key(key) == -1){
log_fatal("move key %s error! %s", key.c_str(), s.code().c_str());
exit(1);
}
}
return (int)keys.size();
}
/**
* Outline of the delete process:
* 1. Initialize the client for this thread if there is no client. This is for the worker
* threads that are attached to any of the threads servicing client requests.
* 2. Grant this thread authorization to perform deletes.
* 3. Temporarily enable mode to bypass shard version checks. TODO: Replace this hack.
* 4. Setup callback to save deletes to moveChunk directory (only if moveParanoia is true).
* 5. Delete range.
* 6. Wait until the majority of the secondaries catch up.
*/
bool RangeDeleterDBEnv::deleteRange(OperationContext* txn,
const RangeDeleteEntry& taskDetails,
long long int* deletedDocs,
std::string* errMsg) {
const string ns(taskDetails.options.range.ns);
const BSONObj inclusiveLower(taskDetails.options.range.minKey);
const BSONObj exclusiveUpper(taskDetails.options.range.maxKey);
const BSONObj keyPattern(taskDetails.options.range.keyPattern);
const WriteConcernOptions writeConcern(taskDetails.options.writeConcern);
const bool fromMigrate = taskDetails.options.fromMigrate;
const bool onlyRemoveOrphans = taskDetails.options.onlyRemoveOrphanedDocs;
const bool initiallyHaveClient = haveClient();
if (!initiallyHaveClient) {
Client::initThread("RangeDeleter");
}
*deletedDocs = 0;
ShardForceVersionOkModeBlock forceVersion;
{
Helpers::RemoveSaver removeSaver("moveChunk",
ns,
taskDetails.options.removeSaverReason);
Helpers::RemoveSaver* removeSaverPtr = NULL;
if (serverGlobalParams.moveParanoia &&
!taskDetails.options.removeSaverReason.empty()) {
removeSaverPtr = &removeSaver;
}
// log the opId so the user can use it to cancel the delete using killOp.
unsigned int opId = txn->getCurOp()->opNum();
log() << "Deleter starting delete for: " << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< ", with opId: " << opId
<< endl;
try {
*deletedDocs =
Helpers::removeRange(txn,
KeyRange(ns,
inclusiveLower,
exclusiveUpper,
keyPattern),
false, /*maxInclusive*/
writeConcern,
removeSaverPtr,
fromMigrate,
onlyRemoveOrphans);
if (*deletedDocs < 0) {
*errMsg = "collection or index dropped before data could be cleaned";
warning() << *errMsg << endl;
if (!initiallyHaveClient) {
txn->getClient()->shutdown();
}
return false;
}
log() << "rangeDeleter deleted " << *deletedDocs
<< " documents for " << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< endl;
}
catch (const DBException& ex) {
*errMsg = str::stream() << "Error encountered while deleting range: "
<< "ns" << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< ", cause by:" << causedBy(ex);
if (!initiallyHaveClient) {
txn->getClient()->shutdown();
}
return false;
}
}
if (!initiallyHaveClient) {
txn->getClient()->shutdown();
}
return true;
}
/**
* Outline of the delete process:
* 1. Initialize the client for this thread if there is no client. This is for the worker
* threads that are attached to any of the threads servicing client requests.
* 2. Grant this thread authorization to perform deletes.
* 3. Temporarily enable mode to bypass shard version checks. TODO: Replace this hack.
* 4. Setup callback to save deletes to moveChunk directory (only if moveParanoia is true).
* 5. Delete range.
* 6. Wait until the majority of the secondaries catch up.
*/
bool RangeDeleterDBEnv::deleteRange(const StringData& ns,
const BSONObj& inclusiveLower,
const BSONObj& exclusiveUpper,
const BSONObj& keyPattern,
bool secondaryThrottle,
std::string* errMsg) {
const bool initiallyHaveClient = haveClient();
if (!initiallyHaveClient) {
Client::initThread("RangeDeleter");
}
ShardForceVersionOkModeBlock forceVersion;
{
Helpers::RemoveSaver removeSaver("moveChunk", ns.toString(), "post-cleanup");
// log the opId so the user can use it to cancel the delete using killOp.
unsigned int opId = cc().curop()->opNum();
log() << "Deleter starting delete for: " << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< ", with opId: " << opId
<< endl;
try {
long long numDeleted =
Helpers::removeRange(KeyRange(ns.toString(),
inclusiveLower,
exclusiveUpper,
keyPattern),
false, /*maxInclusive*/
replSet? secondaryThrottle : false,
serverGlobalParams.moveParanoia ? &removeSaver : NULL,
true, /*fromMigrate*/
true); /*onlyRemoveOrphans*/
if (numDeleted < 0) {
warning() << "collection or index dropped "
<< "before data could be cleaned" << endl;
if (!initiallyHaveClient) {
cc().shutdown();
}
return false;
}
log() << "rangeDeleter deleted " << numDeleted
<< " documents for " << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< endl;
}
catch (const DBException& ex) {
*errMsg = str::stream() << "Error encountered while deleting range: "
<< "ns" << ns
<< " from " << inclusiveLower
<< " -> " << exclusiveUpper
<< ", cause by:" << causedBy(ex);
if (!initiallyHaveClient) {
cc().shutdown();
}
return false;
}
}
if (replSet) {
Timer elapsedTime;
ReplTime lastOpApplied = cc().getLastOp().asDate();
while (!opReplicatedEnough(lastOpApplied,
BSON("w" << "majority").firstElement())) {
if (elapsedTime.seconds() >= 3600) {
*errMsg = str::stream() << "moveChunk repl sync timed out after "
<< elapsedTime.seconds() << " seconds";
if (!initiallyHaveClient) {
cc().shutdown();
}
return false;
}
sleepsecs(1);
}
LOG(elapsedTime.seconds() < 30 ? 1 : 0)
<< "moveChunk repl sync took "
<< elapsedTime.seconds() << " seconds" << endl;
}
if (!initiallyHaveClient) {
cc().shutdown();
}
return true;
}
