Hawkin
/
arangodb
mirror of https://gitee.com/bigwinds/arangodb
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

better error reporting

This commit is contained in:
jsteemann 2016-12-19 14:45:29 +01:00
parent 1132d0820e
commit 818ffbdc49
2 changed files with 43 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
arangod/StorageEngine/MMFilesEngine.cpp
View File

@ -1758,6 +1758,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
std::vector<TRI_datafile_t*> journals;
std::vector<TRI_datafile_t*> sealed;
bool stop = false;
int result = TRI_ERROR_NO_ERROR;
TRI_ASSERT(collection->cid() != 0);
@ -1842,6 +1843,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
LOG_TOPIC(ERR, Logger::DATAFILES)
<< "unable to rename compaction file '" << filename << "' to '"
<< newName << "'";
result = res;
stop = true;
break;
}
@ -1851,6 +1853,8 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
filename = newName;
}
TRI_set_errno(TRI_ERROR_NO_ERROR);
std::unique_ptr<TRI_datafile_t> df(TRI_datafile_t::open(filename, ignoreErrors));
if (df == nullptr) {
@ -1858,6 +1862,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
<< filename
<< "': " << TRI_last_error();
result = TRI_errno();
stop = true;
break;
}
@ -1878,7 +1883,8 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
LOG(ERR) << "collection header mismatch in file '" << filename
<< "', expected TRI_DF_MARKER_COL_HEADER, found "
<< cm->base.getType();
result = TRI_ERROR_ARANGO_CORRUPTED_DATAFILE;
stop = true;
break;
}
@ -1887,6 +1893,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
LOG(ERR) << "collection identifier mismatch, expected "
<< collection->cid() << ", found " << cm->_cid;
result = TRI_ERROR_ARANGO_CORRUPTED_DATAFILE;
stop = true;
break;
}
@ -1918,6 +1925,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
LOG_TOPIC(ERR, Logger::DATAFILES)
<< "datafile '" << filename
<< "' is not sealed, this should never happen";
result = TRI_ERROR_ARANGO_CORRUPTED_DATAFILE;
stop = true;
break;
} else {
@ -1946,6 +1954,7 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
datafiles.emplace_back(datafile);
LOG(DEBUG) << "renamed sealed journal to '" << filename << "'";
} else {
result = res;
stop = true;
LOG(ERR) << "cannot rename sealed log-file to " << filename
<< ", this should not happen: " << TRI_errno_string(res);
@ -1961,6 +1970,9 @@ int MMFilesEngine::openCollection(TRI_vocbase_t* vocbase, LogicalCollection* col
delete datafile;
}
if (result != TRI_ERROR_NO_ERROR) {
return result;
}
return TRI_ERROR_INTERNAL;
}

58
arangod/VocBase/datafile.cpp
View File

@ -1316,35 +1316,37 @@ bool TRI_datafile_t::check(bool ignoreFailures) {
}
}
if (ignoreFailures) {
return fix(currentSize);
if (!ignoreFailures) {
_lastError = TRI_set_errno(TRI_ERROR_ARANGO_CORRUPTED_DATAFILE);
_currentSize = currentSize;
_next = _data + _currentSize;
_state = TRI_DF_STATE_OPEN_ERROR;
LOG(WARN) << "crc mismatch found in datafile '" << getName() << "' of size "
<< _maximalSize << ", at position " << currentSize;
LOG(WARN) << "crc mismatch found inside marker of type '" << TRI_NameMarkerDatafile(marker)
<< "' and size " << size
<< ". expected crc: " << CalculateCrcValue(marker) << ", actual crc: " << marker->getCrc();
if (lastGood != nullptr) {
LOG(INFO) << "last good marker found at: " << hexValue(static_cast<uint64_t>(static_cast<uintptr_t>(lastGood - _data)));
}
printMarker(marker, size, _data, end);
if (nextMarkerOk) {
LOG(INFO) << "data directly following this marker looks ok so repairing the marker manually may recover it...";
LOG(INFO) << "to truncate the file at this marker, please restart the server with the parameter '--wal.ignore-logfile-errors true' if the error happening during WAL recovery, or with parameter '--database.ignore-datafile-errors true' if it happened after WAL recovery";
} else {
LOG(WARN) << "data directly following this marker cannot be analyzed";
}
return false;
}
_lastError = TRI_set_errno(TRI_ERROR_ARANGO_CORRUPTED_DATAFILE);
_currentSize = currentSize;
_next = _data + _currentSize;
_state = TRI_DF_STATE_OPEN_ERROR;
LOG(WARN) << "crc mismatch found in datafile '" << getName() << "' of size "
<< _maximalSize << ", at position " << currentSize;
LOG(WARN) << "crc mismatch found inside marker of type '" << TRI_NameMarkerDatafile(marker)
<< "' and size " << size
<< ". expected crc: " << CalculateCrcValue(marker) << ", actual crc: " << marker->getCrc();
if (lastGood != nullptr) {
LOG(INFO) << "last good marker found at: " << hexValue(static_cast<uint64_t>(static_cast<uintptr_t>(lastGood - _data)));
}
printMarker(marker, size, _data, end);
if (nextMarkerOk) {
LOG(INFO) << "data directly following this marker looks ok so repairing the marker may recover it";
LOG(INFO) << "please restart the server with the parameter '--wal.ignore-logfile-errors true' to repair the marker";
} else {
LOG(WARN) << "data directly following this marker cannot be analyzed";
}
return false;
// ignore failures...
// truncate
return fix(currentSize);
}
}