Reputation: 3179
Is it safe to call write
on Java FileOutputStream
object form multiple threads? Will the output be serialized correctly?
clarification:
In my case the class logger holds a FileOutputStream reference, and multiple threads can call logger write, that formats the output and calls FileOutputStream write.
Should I synchronize my logger write method to warrant that the messages from multiple threads are not mixed?
Upvotes: 12
Views: 27844
Reputation: 12623
No. Java does not support streaming to the same stream from multiple threads.
If you want to do use threaded streams, check out this site: http://lifeinide.com/post/2011-05-25-threaded-iostreams-in-java/
He explains things well and has some sample code for a ThreadedOutputStream
, which would do what you want.
Upvotes: 2
Reputation: 12790
Here is a simple implementation of a synchronized logger using the java nio FileChannel. In this example, log messages are limited to 1024 bytes. You can adjust the log message length by changing the BUFFER_SIZE value.
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.nio.ByteBuffer;
import java.nio.channels.FileChannel;
import java.util.HashMap;
/**
* The MyLogger class abstracts the writing of log messages to a file.
* This is a synchronized implementation due to the usage of java.nio.channels.FileChannel
* which is used to write log messages to the log file.
*
* The MyLogger class maintains a HashMap of MyLogger instances per log file.
* The Key is the MD5 hash of the log file path and the Value is the MyLogger instance for that log file.
*
*/
public final class MyLogger {
private static final int BUFFER_SIZE = 1024;
private static final int DIGEST_BASE_RADIX = 16;
private static final String LINE_SEPARATOR = System.getProperty("line.separator");
private static HashMap<String, MyLogger> sLoggerMap;
private FileChannel mLogOutputChannel;
private ByteBuffer mByteBuffer;
private String mLogDir;
private String mLogFileName;
/**
* Private constructor which creates our log dir and log file if they do not already already exist.
* If the log file exists, then it is opened in append mode.
*
* @param logDir
* The dir where the log file resides
* @param logFileName
* The file name of the log file
* @throws IOException
* Thrown if the file could not be created or opened for writing.
*/
private MyLogger(String logDir, String logFileName) throws IOException {
mLogDir = logDir;
mLogFileName = logFileName;
// create the log dir and log file if they do not exist
FileOutputStream logFile;
new File(mLogDir).mkdirs();
final String logFilePath = mLogDir + File.separatorChar + mLogFileName;
final File f = new File(logFilePath);
if(!f.exists()) {
f.createNewFile();
}
logFile = new FileOutputStream(logFilePath, true);
// set up our output channel and byte buffer
mLogOutputChannel = logFile.getChannel();
mByteBuffer = ByteBuffer.allocate(BUFFER_SIZE);
}
/**
* Writes the given log message to the log file that is represented by this MyLogger instance.
* If the log message could not be written to the log file an error is logged in the System log.
*
* @param logMessage
* The log message to write to the log file.
*/
public void log(String logMessage) {
// write the log message to the log file
if (mLogOutputChannel != null) {
mByteBuffer.put(logMessage.getBytes());
mByteBuffer.put(LINE_SEPARATOR.getBytes());
mByteBuffer.flip();
try {
mLogOutputChannel.write(mByteBuffer);
// ensure that the data we just wrote to the log file is pushed to the disk right away
mLogOutputChannel.force(true);
} catch (IOException e) {
// Could not write to log file output channel
e.printStackTrace();
return;
}
}
if(mByteBuffer != null) {
mByteBuffer.clear();
}
}
/**
* Get an instance of the MyLogger for the given log file. Passing in the same logDir and logFileName will result in the same MyLogger instance being returned.
*
* @param logDir
* The directory path where the log file resides. Cannot be empty or null.
* @param logFileName
* The name of the log file Cannot be empty or null.
* @return The instance of the MyLogger representing the given log file. Null is returned if either logDir or logFilename is null or empty string.
* @throws IOException
* Thrown if the file could not be created or opened for writing.
*/
public static MyLogger getLog(String logDir, String logFileName) throws IOException {
if(logDir == null || logFileName == null || logDir.isEmpty() || logFileName.isEmpty()) {
return null;
}
if(sLoggerMap == null) {
sLoggerMap = new HashMap<String, MyLogger>();
}
final String logFilePathHash = getHash(logDir + File.separatorChar + logFileName);
if(!sLoggerMap.containsKey(logFilePathHash)) {
sLoggerMap.put(logFilePathHash, new MyLogger(logDir, logFileName));
}
return sLoggerMap.get(logFilePathHash);
}
/**
* Utility method for generating an MD5 hash from the given string.
*
* @param path
* The file path to our log file
* @return An MD5 hash of the log file path. If an MD5 hash could not be generated, the path string is returned.
*/
private static String getHash(String path) {
try {
final MessageDigest digest = MessageDigest.getInstance("MD5");
digest.update(path.getBytes());
return new BigInteger(digest.digest()).toString(DIGEST_BASE_RADIX);
} catch (NoSuchAlgorithmException ex) {
// this should never happen, but just to make sure return the path string
return path;
}
}
}
This is how you would use it :
MyLogger aLogger = MyLogger.getLog("/path/to/log/dir", "logFilename");
aLogger.log("my log message");
Upvotes: 5
Reputation: 21459
A file can not be opened more than once in write-mode
, so the answer is no.
After seeing your edit, yes you should introduce synchronization into your logger to make sure the stream is accessed only by one thread at a time. Just a suggestion, why don't you go for Log4J? It already handles your use case.
Upvotes: 8
Reputation: 5681
If you want to keep ordering (ie message 1 in the output stream came before message 2) you have to lock the stream. This in turn reduces concurrency. (All threads will be enqueued in the lock's/semaphore's queue and wait there for the stream to become available to them)
If you're interested only in writing to a stream concurrently and don't care about ordering, you can have buffers for each thread. Each thread writes to its own buffer. When the buffer is full it acquires a lock (which may involve waiting for the lock) on the stream and empties its contents into the stream.
Edit: I just realized that, if you care about ordering and still want multi-threading, if you also write the time in the output stream in unix format (as a long). After the stream is flushed onto some other container, the contents can be sorted based on time and you should have an ordered file.
Upvotes: 1