In my application, I build a Guava Cache object by CacheBuilder.newBuilder() method, and now I need to dynamically adjust some initialization parameters for it.
As I don't find any rebuild method for a guava cache, I have to rebuild a new one.
My question is :
Anybody teach me how to release the old one ? I don't find any useful method either.I just call cache.invalidateAll() for the old one to invalidate all the keys. Is there any risk for OOM ?
As the cache maybe used in multi-threads, is it necessary to declare the cache as volatile ?
my codes is as belows:
private volatile LoadingCache<Long, String> cache = null;
private volatile LoadingCache<Long, String> oldCache = null;
public void rebuildCache(int cacheSize, int expireSeconds) {
logger.info("rebuildCache start: cacheSize: {}, expireSeconds: {}", cacheSize, expireSeconds);
oldCache = cache;
cache = CacheBuilder.newBuilder()
.maximumSize(cacheSize)
.recordStats()
.expireAfterWrite(expireSeconds, TimeUnit.SECONDS)
.build(
new CacheLoader<Long, String>() {
#Override
public String load(Long id) {
// some codes here
}
}
);
if (oldCache != null) {
oldCache.invalidateAll();
}
logger.info("rebuildCache end");
}
public String getByCache(Long id) throws ExecutionException {
return cache.get(id);
}
You don't need to do anything special to release the old one; it'll get garbage collected like any other object. You probably should mark the cache as volatile, or even better, an AtomicReference so multiple threads don't replace the cache at the same time. That said, oldCache should be a variable inside the method, not the class.
I am working on measuing my application metrics using below class in which I increment and decrement metrics.
public class AppMetrics {
private final AtomicLongMap<String> metricCounter = AtomicLongMap.create();
private static class Holder {
private static final AppMetrics INSTANCE = new AppMetrics();
}
public static AppMetrics getInstance() {
return Holder.INSTANCE;
}
private AppMetrics() {}
public void increment(String name) {
metricCounter.getAndIncrement(name);
}
public AtomicLongMap<String> getMetricCounter() {
return metricCounter;
}
}
I am calling increment method of AppMetrics class from multithreaded code to increment the metrics by passing the metric name.
Problem Statement:
Now I want to have metricCounter for each clientId which is a String. That means we can also get same clientId multiple times and sometimes it will be a new clientId, so somehow then I need to extract the metricCounter map for that clientId and increment metrics on that particular map (which is what I am not sure how to do that).
What is the right way to do that keeping in mind it has to be thread safe and have to perform atomic operations. I was thinking to make a map like that instead:
private final Map<String, AtomicLongMap<String>> clientIdMetricCounterHolder = Maps.newConcurrentMap();
Is this the right way? If yes then how can I populate this map by passing clientId as it's key and it's value will be the counter map for each metric.
I am on Java 7.
If you use a map then you'll need to synchronize on the creation of new AtomicLongMap instances. I would recommend using a LoadingCache instead. You might not end up using any of the actual "caching" features but the "loading" feature is extremely helpful as it will synchronizing creation of AtomicLongMap instances for you. e.g.:
LoadingCache<String, AtomicLongMap<String>> clientIdMetricCounterCache =
CacheBuilder.newBuilder().build(new CacheLoader<String, AtomicLongMap<String>>() {
#Override
public AtomicLongMap<String> load(String key) throws Exception {
return AtomicLongMap.create();
}
});
Now you can safely start update metric counts for any client without worrying about whether the client is new or not. e.g.
clientIdMetricCounterCache.get(clientId).incrementAndGet(metricName);
A Map<String, Map<String, T>> is just a Map<Pair<String, String>, T> in disguise. Create a MultiKey class:
class MultiKey {
public String clientId;
public String name;
// be sure to add hashCode and equals
}
Then just use an AtomicLongMap<MultiKey>.
Edited:
Provided the set of metrics is well defined, it wouldn't be too hard to use this data structure to view metrics for one client:
Set<String> possibleMetrics = // all the possible values for "name"
Map<String, Long> getMetricsForClient(String client) {
return Maps.asMap(possibleMetrics, m -> metrics.get(new MultiKey(client, m));
}
The returned map will be a live unmodifiable view. It might be a bit more verbose if you're using an older Java version, but it's still possible.
I have two components:
The manager, on which add(Data) can be called. This will add some data to the manager.
The clients, which can call retrieve(predicate) on the manager. A list of Data objects which match the given predicate are returned. If there is no such data, retrieve keeps waiting.
A typical blocking priority queue cannot be used here, since the client is not interested in every new object. Only those who are allowed by his requirements as defined in the predicate are useful for him.
How can this be implemented in Java? I could get it working with a x.notifyAll() call after each call to add(Data) in the manager, and a x.wait() in the retrieve(predicates) method. I was wondering if the java.concurrent package has more higher-level functionalities which can be used for this problem.
Here is an outline of something that may give you an idea. For simplicity I am going to assume that predicates and data are strings.
As you stated you do not know your predicates ahead of time so I would try to dynamically update and cache based on new incoming predicates.
Manager
public class Manager(){
private Map<String, Set<String>> jobs = new HashMap<>():
private Set<String> knownPredicates = new HasSet();
private final static String GENERAL = "GENERAL_DATA";
public void addJob(String data){
Set<String> matchingPredicates = getMatchingPredicates(data);
if(matchingPredicates.isEmpty()){
updateJobs(GENERAL, data);
} else {
for(String predicate: matchingPredicates){
updateJobs(GENERAL, data);
}
}
synchronized(this){
notifyAll();
}
}
private Set<String> getMatchingPredicates(String data){
Set<String> matchingPredicates = new HashSet<>();
for(String knownPredicate: knownPredicates){
// Check if the data matched the predicate. If so add it to the list
}
return matchingPredicates;
}
private void updateJobs(String predicate, String data){
Set<String> dataList;
if(jobs.containsKey(predicate)){
dataList = jobs.get(predicate);
} else {
dataList = new HashSet<>();
}
dataList.add(data);
jobs.put(predicate, dataList);
}
public synchronized List<String> retrieve(String predicate){
Set<String> jobsToReturn;
knownPredicates.add(predicate);
if(jobs.containsKey(predicate)){
jobsToReturn = jobs.remove(predicate);
}
for(String unknownData: jobs.get()){
//Check if unknownData matches the new predicate if it does add it to jobsToReturn
}
cleanupData(jobsToReturn);
return jobsToReturn;
}
//Removes data that may match more than one predicate
private static void cleanupData(Set<String> dataSet){
for(String data: dataSet){
for(Set <String> predicateSet: jobs.values()){
predicateSet.remove(data);
}
}
}
}
Client
public class Client() implements Runnable{
private Manager managerRef;
public Client(Manager m){
managerRef = m;
}
public void run() {
while(true){
String predicate = //Get the predicate somehow
Set<String> workToDo = managerRef.retrieve(predicate)
if(workToDo.isEmpty()){
synchornized(managerRef){
managerRef.wait();
}
} else {
//Do something
}
}
}
}
The above is only a skeleton though. You would have to resolve some issue regarding clearing your known predicates etc. . .
You might need to consider implementing predicate-based caching with the following behavior:
If 'retrieve(predicate)' method has never been called and 'add(Data)' method is executed, a new Data object is simply added to the manager and cache remains empty.
If 'retrieve(predicate)' method is called, the client checks the cache for the requested predicate in order to retrieve references to the corresponding Data objects. If cache is empty or no match has been found, the system runs a search on the specified predicate against all Data objects in the manager and updates the cache. To improve the performance, if no match found, flag this up in the cache so that the subsequent queries for the same predicate are returned faster.
If 'add(Data)' method is called and cache isn't empty, the Data object being added is scanned for all predicates already in the cache and the matching objects are associated by a reference with the corresponding predicates in the cache.
Note as any caching mechanism, it will be slower at the start but will improve as more objects fill up the cache.
In the book "Java Concurrency in Practice" is mentioned that the following code is not threadsafe:
#NotThreadSafe
public class DoubleCheckedLocking {
private static Resource resource;
public static Resource getInstance(){
if(resource == null){
synchronized (DoubleCheckedLocking.class){
if(resource == null)
resource = new Resource();
}
}
return resource;
}
}
It is not thread safe because because:
- one thread can create new instance of Resource
- another thread at the same time in the "if" condition can get not empty reference but the object of Resource will not be completly initialized
In this question is similar code. Resources are stored in concurentHashMap and people say that it is threadSafe. Something like this:
public class DoubleCheckedLocking2 {
private static ConcurrentHashMap<String, ComplexObject> cache = new ConcurrentHashMap<String, ComplexObject>();
public static ComplexObject getInstance(String key) {
ComplexObject result = cache.get(key);
if (result == null) {
synchronized (DoubleCheckedLocking2.class) {
ComplexObject currentValue = cache.get(key);
if (currentValue == null) {
result = new ComplexObject();
cache.put(key, result);
} else {
result = currentValue;
}
}
}
return result;
}
}
Why does storing the values in ConcurrentHashMap make the code threadSafe? I think that it is still possible that the ComplexObject won't be completely initialized and this "partial object" will be saved in the map. And other threads will be reading partial not fully initialized objects.
I think I know what is "happens-before", I've analyzed code in JDK 8.0_31 and I still don't know the answer.
I am aware of the functions like computeIfAbsent, putIfAbsent. I know that this code can be written differently. I just wan't know details which make this code threadsafe.
Happens before actually is the key here. There's a happens before edge extending from map.put(key, object) to a subsequent map.get(key), therefore the object you retrieve is at least as up to date as it was at the time it was stored in the map.
let's say we have a CountryList object in our application that should return the list of countries. The loading of countries is a heavy operation, so the list should be cached.
Additional requirements:
CountryList should be thread-safe
CountryList should load lazy (only on demand)
CountryList should support the invalidation of the cache
CountryList should be optimized considering that the cache will be invalidated very rarely
I came up with the following solution:
public class CountryList {
private static final Object ONE = new Integer(1);
// MapMaker is from Google Collections Library
private Map<Object, List<String>> cache = new MapMaker()
.initialCapacity(1)
.makeComputingMap(
new Function<Object, List<String>>() {
#Override
public List<String> apply(Object from) {
return loadCountryList();
}
});
private List<String> loadCountryList() {
// HEAVY OPERATION TO LOAD DATA
}
public List<String> list() {
return cache.get(ONE);
}
public void invalidateCache() {
cache.remove(ONE);
}
}
What do you think about it? Do you see something bad about it? Is there other way to do it? How can i make it better? Should i look for totally another solution in this cases?
Thanks.
google collections actually supplies just the thing for just this sort of thing: Supplier
Your code would be something like:
private Supplier<List<String>> supplier = new Supplier<List<String>>(){
public List<String> get(){
return loadCountryList();
}
};
// volatile reference so that changes are published correctly see invalidate()
private volatile Supplier<List<String>> memorized = Suppliers.memoize(supplier);
public List<String> list(){
return memorized.get();
}
public void invalidate(){
memorized = Suppliers.memoize(supplier);
}
Thanks you all guys, especially to user "gid" who gave the idea.
My target was to optimize the performance for the get() operation considering the invalidate() operation will be called very rare.
I wrote a testing class that starts 16 threads, each calling get()-Operation one million times. With this class I profiled some implementation on my 2-core maschine.
Testing results
Implementation Time
no synchronisation 0,6 sec
normal synchronisation 7,5 sec
with MapMaker 26,3 sec
with Suppliers.memoize 8,2 sec
with optimized memoize 1,5 sec
1) "No synchronisation" is not thread-safe, but gives us the best performance that we can compare to.
#Override
public List<String> list() {
if (cache == null) {
cache = loadCountryList();
}
return cache;
}
#Override
public void invalidateCache() {
cache = null;
}
2) "Normal synchronisation" - pretty good performace, standard no-brainer implementation
#Override
public synchronized List<String> list() {
if (cache == null) {
cache = loadCountryList();
}
return cache;
}
#Override
public synchronized void invalidateCache() {
cache = null;
}
3) "with MapMaker" - very poor performance.
See my question at the top for the code.
4) "with Suppliers.memoize" - good performance. But as the performance the same "Normal synchronisation" we need to optimize it or just use the "Normal synchronisation".
See the answer of the user "gid" for code.
5) "with optimized memoize" - the performnce comparable to "no sync"-implementation, but thread-safe one. This is the one we need.
The cache-class itself:
(The Supplier interfaces used here is from Google Collections Library and it has just one method get(). see http://google-collections.googlecode.com/svn/trunk/javadoc/com/google/common/base/Supplier.html)
public class LazyCache<T> implements Supplier<T> {
private final Supplier<T> supplier;
private volatile Supplier<T> cache;
public LazyCache(Supplier<T> supplier) {
this.supplier = supplier;
reset();
}
private void reset() {
cache = new MemoizingSupplier<T>(supplier);
}
#Override
public T get() {
return cache.get();
}
public void invalidate() {
reset();
}
private static class MemoizingSupplier<T> implements Supplier<T> {
final Supplier<T> delegate;
volatile T value;
MemoizingSupplier(Supplier<T> delegate) {
this.delegate = delegate;
}
#Override
public T get() {
if (value == null) {
synchronized (this) {
if (value == null) {
value = delegate.get();
}
}
}
return value;
}
}
}
Example use:
public class BetterMemoizeCountryList implements ICountryList {
LazyCache<List<String>> cache = new LazyCache<List<String>>(new Supplier<List<String>>(){
#Override
public List<String> get() {
return loadCountryList();
}
});
#Override
public List<String> list(){
return cache.get();
}
#Override
public void invalidateCache(){
cache.invalidate();
}
private List<String> loadCountryList() {
// this should normally load a full list from the database,
// but just for this instance we mock it with:
return Arrays.asList("Germany", "Russia", "China");
}
}
Whenever I need to cache something, I like to use the Proxy pattern.
Doing it with this pattern offers separation of concerns. Your original
object can be concerned with lazy loading. Your proxy (or guardian) object
can be responsible for validation of the cache.
In detail:
Define an object CountryList class which is thread-safe, preferably using synchronization blocks or other semaphore locks.
Extract this class's interface into a CountryQueryable interface.
Define another object, CountryListProxy, that implements the CountryQueryable.
Only allow the CountryListProxy to be instantiated, and only allow it to be referenced
through its interface.
From here, you can insert your cache invalidation strategy into the proxy object. Save the time of the last load, and upon the next request to see the data, compare the current time to the cache time. Define a tolerance level, where, if too much time has passed, the data is reloaded.
As far as Lazy Load, refer here.
Now for some good down-home sample code:
public interface CountryQueryable {
public void operationA();
public String operationB();
}
public class CountryList implements CountryQueryable {
private boolean loaded;
public CountryList() {
loaded = false;
}
//This particular operation might be able to function without
//the extra loading.
#Override
public void operationA() {
//Do whatever.
}
//This operation may need to load the extra stuff.
#Override
public String operationB() {
if (!loaded) {
load();
loaded = true;
}
//Do whatever.
return whatever;
}
private void load() {
//Do the loading of the Lazy load here.
}
}
public class CountryListProxy implements CountryQueryable {
//In accordance with the Proxy pattern, we hide the target
//instance inside of our Proxy instance.
private CountryQueryable actualList;
//Keep track of the lazy time we cached.
private long lastCached;
//Define a tolerance time, 2000 milliseconds, before refreshing
//the cache.
private static final long TOLERANCE = 2000L;
public CountryListProxy() {
//You might even retrieve this object from a Registry.
actualList = new CountryList();
//Initialize it to something stupid.
lastCached = Long.MIN_VALUE;
}
#Override
public synchronized void operationA() {
if ((System.getCurrentTimeMillis() - lastCached) > TOLERANCE) {
//Refresh the cache.
lastCached = System.getCurrentTimeMillis();
} else {
//Cache is okay.
}
}
#Override
public synchronized String operationB() {
if ((System.getCurrentTimeMillis() - lastCached) > TOLERANCE) {
//Refresh the cache.
lastCached = System.getCurrentTimeMillis();
} else {
//Cache is okay.
}
return whatever;
}
}
public class Client {
public static void main(String[] args) {
CountryQueryable queryable = new CountryListProxy();
//Do your thing.
}
}
Your needs seem pretty simple here. The use of MapMaker makes the implementation more complicated than it has to be. The whole double-checked locking idiom is tricky to get right, and only works on 1.5+. And to be honest, it's breaking one of the most important rules of programming:
Premature optimization is the root of
all evil.
The double-checked locking idiom tries to avoid the cost of synchronization in the case where the cache is already loaded. But is that overhead really causing problems? Is it worth the cost of more complex code? I say assume it is not until profiling tells you otherwise.
Here's a very simple solution that requires no 3rd party code (ignoring the JCIP annotation). It does make the assumption that an empty list means the cache hasn't been loaded yet. It also prevents the contents of the country list from escaping to client code that could potentially modify the returned list. If this is not a concern for you, you could remove the call to Collections.unmodifiedList().
public class CountryList {
#GuardedBy("cache")
private final List<String> cache = new ArrayList<String>();
private List<String> loadCountryList() {
// HEAVY OPERATION TO LOAD DATA
}
public List<String> list() {
synchronized (cache) {
if( cache.isEmpty() ) {
cache.addAll(loadCountryList());
}
return Collections.unmodifiableList(cache);
}
}
public void invalidateCache() {
synchronized (cache) {
cache.clear();
}
}
}
I'm not sure what the map is for. When I need a lazy, cached object, I usually do it like this:
public class CountryList
{
private static List<Country> countryList;
public static synchronized List<Country> get()
{
if (countryList==null)
countryList=load();
return countryList;
}
private static List<Country> load()
{
... whatever ...
}
public static synchronized void forget()
{
countryList=null;
}
}
I think this is similar to what you're doing but a little simpler. If you have a need for the map and the ONE that you've simplified away for the question, okay.
If you want it thread-safe, you should synchronize the get and the forget.
What do you think about it? Do you see something bad about it?
Bleah - you are using a complex data structure, MapMaker, with several features (map access, concurrency-friendly access, deferred construction of values, etc) because of a single feature you are after (deferred creation of a single construction-expensive object).
While reusing code is a good goal, this approach adds additional overhead and complexity. In addition, it misleads future maintainers when they see a map data structure there into thinking that there's a map of keys/values in there when there is really only 1 thing (list of countries). Simplicity, readability, and clarity are key to future maintainability.
Is there other way to do it? How can i make it better? Should i look for totally another solution in this cases?
Seems like you are after lazy-loading. Look at solutions to other SO lazy-loading questions. For example, this one covers the classic double-check approach (make sure you are using Java 1.5 or later):
How to solve the "Double-Checked Locking is Broken" Declaration in Java?
Rather than just simply repeat the solution code here, I think it is useful to read the discussion about lazy loading via double-check there to grow your knowledge base. (sorry if that comes off as pompous - just trying teach to fish rather than feed blah blah blah ...)
There is a library out there (from atlassian) - one of the util classes called LazyReference. LazyReference is a reference to an object that can be lazily created (on first get). it is guarenteed thread safe, and the init is also guarenteed to only occur once - if two threads calls get() at the same time, one thread will compute, the other thread will block wait.
see a sample code:
final LazyReference<MyObject> ref = new LazyReference() {
protected MyObject create() throws Exception {
// Do some useful object construction here
return new MyObject();
}
};
//thread1
MyObject myObject = ref.get();
//thread2
MyObject myObject = ref.get();
This looks ok to me (I assume MapMaker is from google collections?) Ideally you wouldn't need to use a Map because you don't really have keys but as the implementation is hidden from any callers I don't see this as a big deal.
This is way to simple to use the ComputingMap stuff. You only need a dead simple implementation where all methods are synchronized, and you should be fine. This will obviously block the first thread hitting it (getting it), and any other thread hitting it while the first thread loads the cache (and the same again if anyone calls the invalidateCache thing - where you also should decide whether the invalidateCache should load the cache anew, or just null it out, letting the first attempt at getting it again block), but then all threads should go through nicely.
Use the Initialization on demand holder idiom
public class CountryList {
private CountryList() {}
private static class CountryListHolder {
static final List<Country> INSTANCE = new List<Country>();
}
public static List<Country> getInstance() {
return CountryListHolder.INSTANCE;
}
...
}
Follow up to Mike's solution above. My comment didn't format as expected... :(
Watch out for synchronization issues in operationB, especially since load() is slow:
public String operationB() {
if (!loaded) {
load();
loaded = true;
}
//Do whatever.
return whatever;
}
You could fix it this way:
public String operationB() {
synchronized(loaded) {
if (!loaded) {
load();
loaded = true;
}
}
//Do whatever.
return whatever;
}
Make sure you ALWAYS synchronize on every access to the loaded variable.