I have a question about caching in spring using Caffeine.
I have a cache configuration:
#Configuration
public class CaffeineCacheConfig {
#Bean
public CacheManager cacheManager() {
CaffeineCacheManager cacheManager = new CaffeineCacheManager(
"someDataCache1",
"someDataCache2");
cacheManager.setCaffeine(caffeineCacheBuilder());
return cacheManager;
}
Caffeine<Object, Object> caffeineCacheBuilder() {
return Caffeine.newBuilder()
.initialCapacity(100)
.maximumSize(200)
.expireAfterWrite(1, TimeUnit.DAYS)
.recordStats();
}
}
And a #Cachaeble method:
#Override
#Cacheable
public List<SomeData> getSomeData(String group) {
return someService.getSomeDataForGroup(group);
}
Now, the group values are predefined, they can only be group1, group2 and group3 will be added in the future.
Can this caching mechanism be configured basing on String group value, so someDataCache1 will be used when String group will be group1 and someDataCache1 will be used when String group is group2 etc?
You can add multiple conditional #Cacheable annotations to the #Caching annotation.
#Caching(cacheable = {
#Cacheable(cacheNames = "someDataCache1", condition = "#group.equals('group1')"),
#Cacheable(cacheNames = "someDataCache2", condition = "#group.equals('group2')")
})
public List<String> getSomeData(String group) {
return someService.getSomeDataForGroup(group);
}
Related
I'm trying to use Spring Cache to store data, generated by another method inside Service class.
This method marked with #Cacheable is a public method, the cache is being called in Controller layer.
When I do debugging, I inspect the object cacheManager, I found that it contains the map that I stored, but when calling the method cacheManager.getCache("cache") it return null.
Question is why that method return null while the object is holding the value?
This is the config, service and controller:
Spring bean config:
#EnableCaching
public class CachingConfig {
#Bean
public CacheManager cacheManager() {
return new ConcurrentMapCacheManager("optCache");
}
}
Service:
public void verify(Request request, String authorization) {
String memberId = parseAuthToken(authorization).getMembershipID();
buildOtpCache(memberId, request.getTokenUUID(), 0)
}
#Cacheable("otpCache")
public OTPCache buildOtpCache(String memberId, String uuid, int counter) {
return OTPCache.builder()
.memberId(memberId)
.tokenUUID(uuid)
.timestamp(LocalDateTime.now())
.counter(counter)
.build();
}
Controller:
#Override
public void verifyOTP(MeOTPVerifyRequest verifyOTPRequest, String authorization) {
String memberId = parseAuthToken(authorization).getMembershipID();
Collection<String> a = cacheManager.getCacheNames();
OTPCache OTPCache = cacheManager.getCache("cache").get(memberId, OTPCache.class);
otpService.verify(verifyOTPRequest, authorization);
}
EDIT:
This is my new service class, remove #Cacheable annotation:
public void verify(Request request, String authorization) {
String memberId = parseAuthToken(authorization).getMembershipID();
cacheManager.getCache("optCache").put(memberId, buildOtpCache(memberId, request.getTokenUUID(), 0));
}
public OTPCache buildOtpCache(String memberId, String uuid, int counter) {
return OTPCache.builder()
.memberId(memberId)
.tokenUUID(uuid)
.timestamp(LocalDateTime.now())
.counter(counter)
.build();
}
I'm currently updating from Spring boot 2.2.x to 2.6.x + legacy code, it's a big jump so there were multiple changes. I'm now running into a problem with load balancing through an api-gateway. I'll apologize in advance for the wall of code to come. I will put the point of failure at the bottom.
When I send in an API request, I get the following error:
more than one 'primary' bean found among candidates: [zookeeperDiscoveryClientServiceInstanceListSupplier, serviceInstanceListSupplier, retryAwareDiscoveryClientServiceInstanceListSupplier]
it seems that the zookeeperDiscovery and retryAware suppliers are loaded through the default serviceInsatnceListSupplier, which has #Primary over it. I thought would take precedence over the other ones. I assume I must be doing something wrong due changes in the newer version, here are the relevant code in question:
#Configuration
#LoadBalancerClients(defaultConfiguration = ClientConfiguration.class)
public class WebClientConfiguration {
#Bean
#Qualifier("microserviceWebClient")
#ConditionalOnMissingBean(name = "microserviceWebClient")
public WebClient microserviceWebClient(#Qualifier("microserviceWebClientBuilder") WebClient.Builder builder) {
return builder.build();
}
#Bean
#Qualifier("microserviceWebClientBuilder")
#ConditionalOnMissingBean(name = "microserviceWebClientBuilder")
#LoadBalanced
public WebClient.Builder microserviceWebClientBuilder() {
return WebClient.builder();
}
#Bean
#Primary
public ReactorLoadBalancerExchangeFilterFunction reactorLoadBalancerExchangeFilterFunction(
ReactiveLoadBalancer.Factory<ServiceInstance> loadBalancerFactory) {
//the transformer is currently null, there wasn't a transformer before the upgrade
return new CustomExchangeFilterFunction(loadBalancerFactory, transformer);
}
}
There are also some Feign Client related configs here which I will omit, since it's not (or shouldn't be) playing a role in this problem:
public class ClientConfiguration {
/**
* The property key within the feign clients configuration context for the feign client name.
*/
public static final String FEIGN_CLIENT_NAME_PROPERTY = "feign.client.name";
public ClientConfiguration() {
}
//Creates a new BiPredicate for shouldClose. This will be used to determine if HTTP Connections should be automatically closed or not.
#Bean
#ConditionalOnMissingBean
public BiPredicate<Response, Type> shouldClose() {
return (Response response, Type type) -> {
if(type instanceof Class) {
Class<?> currentClass = (Class<?>) type;
return (null == AnnotationUtils.getAnnotation(currentClass, EnableResponseStream.class));
}
return true;
};
}
//Creates a Custom Decoder
#Bean
public Decoder createCustomDecoder(
ObjectFactory<HttpMessageConverters> converters, BiPredicate<Response, Type> shouldClose
) {
return new CustomDecoder(converters, shouldClose);
}
#Bean
#Qualifier("loadBalancerName")
public String loadBalancerName(PropertyResolver propertyResolver) {
String name = propertyResolver.getProperty(FEIGN_CLIENT_NAME_PROPERTY);
if(StringUtils.hasText(name)) {
// we are in a feign context
return name;
}
// we are in a LoadBalancerClientFactory context
name = propertyResolver.getProperty(LoadBalancerClientFactory.PROPERTY_NAME);
Assert.notNull(name, "Could not find a load balancer name within the configuration context!");
return name;
}
#Bean
public ReactorServiceInstanceLoadBalancer reactorServiceInstanceLoadBalancer(
BeanFactory beanFactory, #Qualifier("loadBalancerName") String loadBalancerName
) {
return new CustomRoundRobinLoadBalancer(
beanFactory.getBeanProvider(ServiceInstanceListSupplier.class),
loadBalancerName
);
}
#Bean
#Primary
public ServiceInstanceListSupplier serviceInstanceListSupplier(
#Qualifier(
"filter"
) Predicate<ServiceInstance> filter, DiscoveryClient discoveryClient, Environment environment, #Qualifier(
"loadBalancerName"
) String loadBalancerName
) {
// add service name to environment if necessary
if(environment.getProperty(LoadBalancerClientFactory.PROPERTY_NAME) == null) {
StandardEnvironment wrapped = new StandardEnvironment();
if(environment instanceof ConfigurableEnvironment) {
((ConfigurableEnvironment) environment).getPropertySources()
.forEach(s -> wrapped.getPropertySources().addLast(s));
}
Map<String, Object> additionalProperties = new HashMap<>();
additionalProperties.put(LoadBalancerClientFactory.PROPERTY_NAME, loadBalancerName);
wrapped.getPropertySources().addLast(new MapPropertySource(loadBalancerName, additionalProperties));
environment = wrapped;
}
return new FilteringInstanceListSupplier(filter, discoveryClient, environment);
}
}
There was a change in the ExchangeFilter constructor, but as far as I can tell, it accepts that empty transformer,I don't know if it's supposed to:
public class CustomExchangeFilterFunction extends ReactorLoadBalancerExchangeFilterFunction {
private static final ThreadLocal<ClientRequest> REQUEST_HOLDER = new ThreadLocal<>();
//I think it's wrong but I don't know what to do here
private static List<LoadBalancerClientRequestTransformer> transformersList;
private final Factory<ServiceInstance> loadBalancerFactory;
public CustomExchangeFilterFunction (Factory<ServiceInstance> loadBalancerFactory) {
this(loadBalancerFactory);
///according to docs, but I don't know where and if I need to use this
#Bean
public LoadBalancerClientRequestTransformer transformer() {
return new LoadBalancerClientRequestTransformer() {
#Override
public ClientRequest transformRequest(ClientRequest request, ServiceInstance instance) {
return ClientRequest.from(request)
.header(instance.getInstanceId())
.build();
}
};
}
public CustomExchangeFilterFunction (Factory<ServiceInstance> loadBalancerFactory, List<LoadBalancerClientRequestTransformer> transformersList) {
super(loadBalancerFactory, transformersList); //the changed constructor
this.loadBalancerFactory = loadBalancerFactory;;
}
#Override
public Mono<ClientResponse> filter(ClientRequest request, ExchangeFunction next) {
// put the current request into the thread context - ugly, but couldn't find a better way to access the request within
// the choose method without reimplementing nearly everything
REQUEST_HOLDER.set(request);
try {
return super.filter(request, next);
} finally {
REQUEST_HOLDER.remove();
}
}
//used to be an override, but the function has changed
//code execution doesn't even get this far yet
protected Mono<Response<ServiceInstance>> choose(String serviceId) {
ReactiveLoadBalancer<ServiceInstance> loadBalancer = loadBalancerFactory.getInstance(serviceId);
if(loadBalancer == null) {
return Mono.just(new EmptyResponse());
}
ClientRequest request = REQUEST_HOLDER.get();
// this might be null, if the underlying implementation changed and this method is no longer executed in the same
// thread
// as the filter method
Assert.notNull(request, "request must not be null, underlying implementation seems to have changed");
return choose(loadBalancer, filter);
}
protected Mono<Response<ServiceInstance>> choose(
ReactiveLoadBalancer<ServiceInstance> loadBalancer,
Predicate<ServiceInstance> filter
) {
return Mono.from(loadBalancer.choose(new DefaultRequest<>(filter)));
}
}
There were pretty big changes in the CustomExchangeFilterFunction, but the current execution doesn't even get there. It fails here, in .getIfAvailable(...):
public class CustomRoundRobinLoadBalancer implements ReactorServiceInstanceLoadBalancer {
private static final int DEFAULT_SEED_POSITION = 1000;
private final ObjectProvider<ServiceInstanceListSupplier> serviceInstanceListSupplierProvider;
private final String serviceId;
private final int seedPosition;
private final AtomicInteger position;
private final Map<String, AtomicInteger> positionsForVersions = new HashMap<>();
public CustomRoundRobinLoadBalancer (
ObjectProvider<ServiceInstanceListSupplier> serviceInstanceListSupplierProvider,
String serviceId
) {
this(serviceInstanceListSupplierProvider, serviceId, new Random().nextInt(DEFAULT_SEED_POSITION));
}
public CustomRoundRobinLoadBalancer (
ObjectProvider<ServiceInstanceListSupplier> serviceInstanceListSupplierProvider,
String serviceId,
int seedPosition
) {
Assert.notNull(serviceInstanceListSupplierProvider, "serviceInstanceListSupplierProvider must not be null");
Assert.notNull(serviceId, "serviceId must not be null");
this.serviceInstanceListSupplierProvider = serviceInstanceListSupplierProvider;
this.serviceId = serviceId;
this.seedPosition = seedPosition;
this.position = new AtomicInteger(seedPosition);
}
#Override
// we have no choice but to use the raw type Request here, because this method overrides another one with this signature
public Mono<Response<ServiceInstance>> choose(#SuppressWarnings("rawtypes") Request request) {
//fails here!
ServiceInstanceListSupplier supplier = serviceInstanceListSupplierProvider
.getIfAvailable(NoopServiceInstanceListSupplier::new);
return supplier.get().next().map((List<ServiceInstance> instances) -> getInstanceResponse(instances, request));
}
}
Edit: after some deeper stacktracing, it seems that it does go into the CustomFilterFunction and invokes the constructor with super(loadBalancerFactory, transformer)
I found the problem or a workaround. I was using #LoadBalancerClients because I thought it would just set the same config for all clients that way (even if I technically only have one atm). I changed it to ##LoadBalancerClient and it suddenly worked. I don't quite understand why this made a difference but it did!
I want to cache some db data. for example Cache Customer and use customer.id as the key.
How could I set the key if I want to load all customers (allCustomer() in the code) ?
#Cacheable(value = "customer", key = "#customerID")
public Customer getCustomer(Long customerID) {
return getCustomerData(customerID);
}
// How to setup this key?
#Cacheable(value = "customer", key = "?")
public List<Customer> allCustomer(){
return db.values().stream().collect(Collectors.toList());
}
#CachePut(value = "customer", key = "#customer.id")
public void updateCustomer(Customer customer){
db.put(customer.getId(), customer);
}
#CacheEvict(value = "customer", key = "#customerID")
public void deleteCustomer(Long customerID){
db.remove(customerID);
}
I would recommend using #CachePut instead of #Cacheable. In the case that a new entry is added to the DB from outside of this application instance, the cache would not contain that new value.
You can use #result.id to tell Spring which value to use as a key and I've included a conditional so that you don't get strange errors in case of a null value.
#CachePut(value = "customer", key = "#result.id", condition = "#result != null")
It's impossible to do it for collections with the Spring's annotations - with #Cacheable you'd have just one element in a cache with a computed key and a value with the whole list inside.
If performance is not that important in your app, use getCustomer(...) in a loop.
Otherwise, you'll need to update your cache manually. Unfortunately, Cache interface doesn't provide a method to retrieve all keys/values/key-value pairs from a cache, so a bit of casting is required.
The example for the default in-memory cache (spring.cache.type=simple):
#Autowired
private org.springframework.CacheManager cacheManager;
public List<Customer> allCustomers() {
ConcurrentMap<Long, Customer> customerCache = (ConcurrentMap<Long, Customer>)
cacheManager.getCache("customer").getNativeCache();
if (!customerCache.isEmpty()) {
return new ArrayList<>(customerCache.values());
}
List<Customer> customers = db.values().stream().collect(Collectors.toList());
customers.forEach(customer -> customerCache.put(customer.getId(), customer));
return customers;
}
Or for spring.cache.type=jcache with backed EhCache 3:
#Autowired
private org.springframework.CacheManager cacheManager;
public List<Customer> allCustomers() {
javax.cache.Cache<Long, Customer> customerCache = (javax.cache.Cache<Long, Customer>)
cacheManager.getCache("customer").getNativeCache();
Iterator<Cache.Entry<Long, Customer>> iterator = customerCache.iterator();
List<Customer> cachedCustomers = new ArrayList<>();
while (iterator.hasNext()) {
Cache.Entry<Long, Customer> entry = iterator.next();
cachedCustomers.add(entry.getValue());
}
if (!cachedCustomers.isEmpty()) {
return cachedCustomers;
}
List<Customer> customers = db.values().stream().collect(Collectors.toList());
customers.forEach(customer -> customerCache.put(customer.getId(), customer));
return customers;
}
The same can be done similarly for any other cache type (redis, hazelcast, caffeine etc.).
The corresponding eviction method can be written much easier:
#CacheEvict(value = "customer", allEntries = true)
public void deleteAllCustomers(){
db.removeAll(); //pseudocode
}
I'm looking to add several distinct LoadingCache's to a Spring CacheManager, however I don't see how this is possible using CaffeineCacheManager. It appears that only a single loader is possible for refreshing content, however I need separate loaders for each cache. Is it possible to add multiple loading caches to a Spring cache manager? If so, then how?
CaffeineCacheManager cacheManage = new CaffeineCacheManager();
LoadingCache<String, Optional<Edition>> loadingCache1 =
Caffeine.newBuilder()
.maximumSize(150)
.refreshAfterWrite(5, TimeUnit.MINUTES)
.build(test -> this.testRepo.find(test));
LoadingCache<String, Optional<Edition>> loadingCache2 =
Caffeine.newBuilder()
.maximumSize(150)
.refreshAfterWrite(5, TimeUnit.MINUTES)
.build(test2 -> this.testRepo.find2(test2));
// How do I add to cache manager, and specify a name?
Yes it is possible. Since you need to fine tune every cache, you are probably better at defining them yourself. Back to your example, the next step would be:
SimpleCacheManager cacheManager = new SimpleCacheManager();
cacheManager.setCaches(Arrays.asList(
new CaffeineCache("first", loadingCache1),
new CaffeineCache("second", loadingCache2)));
And then you can use that as usual, e.g.
#Cacheable("first")
public Foo load(String id) { ... }
If you are using Spring Boot, you can just expose the individual cache as beans (so org.springframework.cache.Cache implementations) and we'll detect them and create a SimpleCacheManager automatically for you.
Note that this strategy allows you to use the cache abstraction with different implementations. first could be a caffeine cache and second a cache from another provider.
Having this class will allow you to use #Cacheable("cacheA") where you want as normal:
#EnableCaching
#Configuration
public class CacheConfiguration {
#Bean
public CacheManager cacheManager() {
CaffeineCacheManager manager = new CaffeineCacheManager();
manager.registerCustomCache("cacheA", defaultCache());
manager.registerCustomCache("cacheB", bigCache());
manager.registerCustomCache("cacheC", longCache());
// to avoid dynamic caches and be sure each name is assigned to a specific config (dynamic = false)
// throws error when tries to use a new cache
manager.setCacheNames(Collections.emptyList());
return manager;
}
private static Cache<Object, Object> defaultCache() {
return Caffeine.newBuilder()
.maximumSize(1000)
.expireAfterWrite(5, TimeUnit.MINUTES)
.build();
}
private static Cache<Object, Object> bigCache() {
return Caffeine.newBuilder()
.maximumSize(5000)
.expireAfterWrite(5, TimeUnit.MINUTES)
.build();
}
private static Cache<Object, Object> longCache() {
return Caffeine.newBuilder()
.maximumSize(1000)
.expireAfterWrite(1, TimeUnit.HOURS)
.build();
}
}
Thanks for #rado, this is improved version of his answer. This way we can configure the cache from application properties directly
cache:
specs:
big-cache:
expire-after: WRITE
timeout: 2h
max-size: 1000
long-cache:
expire-after: ACCESS
timeout: 30d
max-size: 100
We need a cache properties for this
#Data
#EnableConfigurationProperties
#Configuration
#ConfigurationProperties(prefix = "cache")
public class CacheProperties {
private static final int DEFAULT_CACHE_SIZE = 100;
private Map<String, CacheSpec> specs = new HashMap<>();
#Data
public static class CacheSpec {
private Duration timeout;
private Integer maxSize = DEFAULT_CACHE_SIZE;
private ExpireAfter expireAfter = ExpireAfter.WRITE;
}
enum ExpireAfter { WRITE, ACCESS }
}
And then we can configure directly from external config file
#EnableCaching
#Configuration
#RequiredArgsConstructor
public class CacheConfiguration {
private final CacheProperties cacheProperties;
#Bean
public CacheManager cacheManager() {
CaffeineCacheManager manager = new CaffeineCacheManager();
Map<String, CacheProperties.CacheSpec> specs = cacheProperties.getSpecs();
specs.keySet().forEach(cacheName -> {
CacheProperties.CacheSpec spec = specs.get(cacheName);
manager.registerCustomCache(cacheName, buildCache(spec));
});
// to avoid dynamic caches and be sure each name is assigned
// throws error when tries to use a new cache
manager.setCacheNames(Collections.emptyList());
return manager;
}
private Cache<Object, Object> buildCache(CacheProperties.CacheSpec cacheSpec) {
if (cacheSpec.getExpireAfter() == CacheProperties.ExpireAfter.ACCESS) {
return Caffeine.newBuilder()
.expireAfterAccess(cacheSpec.getTimeout())
.build();
}
return Caffeine.newBuilder()
.expireAfterWrite(cacheSpec.getTimeout())
.build();
}
}
Now you can use the cache with using cache name
#Cacheable(cacheNames = "big-cache", key = "{#key}", unless="#result == null")
public Object findByKeyFromBigCache(String key) {
// create the required object and return
}
#Cacheable(cacheNames = "long-cache", key = "{#key}", unless="#result == null")
public Object findByKeyFromLongCache(String key) {
// create the required object and return
}
I've got next method in UserService:
#Cacheable(value = "user", key="#p0")
public User find(String user) {
return userRepository.findByUser(User);
}
It caches well. In other service I have:
#Transactional
public void updateToken(int id, String token) {
Group group = groupRepository.findOne(id);
group.getMembers().forEach(member -> {
member.getUser().setToken(token);
removeUserCacheByName(member.getUser().getName());
});
groupRepository.save(group);
}
#CacheEvict(value = "user", key="#p0")
public void removeUserCacheByName(String name) {
log.debug("Removing user cache by name {}.", name);
}
After updateToken method, cache does not clear.
What you're seeing is normal. You're calling removeUserCacheByName() from within the Proxy object so the catching aspect doesn't execute. You have this behaviour explained in the documentation.
You can do some things to work around this:
1) Take the evict method (removeUserCacheByName) to another bean, autowire it in updateToken()'s class, and call it.
2) An ugly but useful one, autowire the ApplicationContext, get the same object from it and call the method, e.g.:
public class UserService{
#Autowired
private ApplicationContext ac;
#Transactional
public void updateToken(int id, String token) {
Group group = groupRepository.findOne(id);
group.getMembers().forEach(member -> {
member.getUser().setToken(token);
UserService sameBean = ac.getBean(UserService.class);
sameBean.removeUserCacheByName(member.getUser().getName());
});
groupRepository.save(group);
}
#CacheEvict(value = "user", key="#p0")
public void removeUserCacheByName(String name) {
log.debug("Removing user cache by name {}.", name);
}
}