Hands-on RedisTemplate
We are improving cache in the project Im working on. In the code base I’ve found a method which accepted a collection of ids and returned collection of objects for those ids. My goal was to add cache on individual values returned by the method. Creation of items was IO heavy and time-consuming operation so removal of cache was not an option.
I’m working with old system and logic for loading and invalidating values was complicated and scattered across many places. To change this serious refactoring would be needed so I’ve decided to leave it be, but to improve codebase a bit and fix cache layer (code of various business operations was mixed with native Ehcache API calls).
I wanted to cache items separately because it was up to the user to decide what elements he wants to access and “up to him to decide” what elements should be invalidated. The main challenge was in the way the cache was used. If you image your method as something like (called from different places and invalidated in other on single item basis):
List<Item> findItems(Collection<UUID> ids)It’s been called from different places like (and always invalidated for single element):
findItems(asList(1,2,3));
findItems(asList(1));
findItems(asList(2,3))My first concept was to implement this caching mechanism using spring’s cache abstraction, but it is based on a pretty simple idea and it does not allow such behavior, even with a custom key generator, all you can do is to convert input arguments to something different, not to mention invalidate single value from cached collection… Spring cache abstraction doesn’t solve the problem for me :( Because of that I had to improvise a bit and solve the issue by myself.
As we are using Redis for caching I’ve decided that this is vendor lock I’m willing to live with, but firstly I’ve created abstraction so there will be only one place to change in case we’ll need to pick different cache vendor and we will not have to worry about implementation details and how to use the cache in the future except this one class responsible for it.
interface ItemsCache {
List<Item> findItems(Collection<UUID> ids);
void saveItems(Collection<Item> items);
}In reality there was more but let’s keep it simple.
Next, all I had to do was implement a basic caching mechanism using spring’s RedisTemplate. I’ve started from writing tests to avoid rebooting application or cleaning up Redis every time I’ve messed something up (serialization of those objects to JSON was not so easy as those were pretty complex and big structures…):
@Test
public void items_are_loaded_from_cache() {
//given
final Item item1 = anyItem();
final Item item2 = anyItem();
Mockito
.when(slowItemRepository.get(item1.getId()))
.thenReturn(item1);
Mockito
.when(slowItemRepository.get(item2.getId()))
.thenReturn(item2);
//when
itemsService.findItems(singleton(item1.getId()));
itemsService.findItems(singleton(item2.getId()));
itemsService.findItems(asList(item1.getId(), item2.getId()));
//then
Mockito.verify(slowItemRepository, times(1)).get(item1.getId());
Mockito.verify(slowItemRepository, times(1)).get(item2.getId());
}
@Test
public void ttl_is_set_on_items() {
//given
final Item item = anyItem();
Mockito
.when(slowItemRepository.get(item.getId()))
.thenReturn(item);
//when
itemsService.findItems(singleton(item.getId()));
//then
long ttl = jedisConnectionFactory.getConnection().pTtl(("Item:" + item.getId()).getBytes());
Assert.assertTrue(ttl > 0);
}The last step was to write actual implementation of the service:
@Service
@RequiredArgsConstructor
public class ItemsService {
private final SlowItemRepository repository;
private final ItemsCache itemsCache;
public List<Item> findItems(Collection<UUID> ids) {
final List<Item> foundInCache = itemsCache.findItems(ids);
final List<Item> notCached = loadMissingItems(ids, foundInCache);
final List<Item> result = Stream
.concat(foundInCache.stream(), notCached.stream())
.collect(toList());
itemsCache.saveItems(notCached);
return result;
}
private List<Item> loadMissingItems(Collection<UUID> ids, List<Item> foundInCache) {
final Set<UUID> cachedIds = foundInCache.stream()
.map(Item::getId)
.collect(Collectors.toSet());
return ids.stream()
.filter(id -> !cachedIds.contains(id))
.map(this::lookupInRepository)
.collect(toList());
}
private Item lookupInRepository(UUID id) {
return repository.get(id);
}
}And caching itself:
@Component
@RequiredArgsConstructor
public class ItemsCacheImpl implements ItemsCache {
private final RedisTemplate<String, Item> redisTemplate;
@Override
public List<Item> findItems(Collection<UUID> ids) {
final List<Item> result = redisTemplate.opsForValue()
.multiGet(ids.stream().map(this::generateCacheKey).collect(toList()))
.stream()
.filter(Objects::nonNull)
.collect(toList());
refreshTTL(result);
return result;
}
@Override
public void saveItems(Collection<Item> items) {
redisTemplate.opsForValue()
.multiSet(items.stream()
.collect(Collectors.toMap(
item -> generateCacheKey(item.getId()),
Function.identity())));
refreshTTL(items);
}
private void refreshTTL(Collection<Item> result) {
redisTemplate.executePipelined((SessionCallback) callback -> {
result.stream()
.map(item -> generateCacheKey(item.getId()))
.forEach(key -> callback.expire(key, 10, TimeUnit.MINUTES));
return null;
});
}
private String generateCacheKey(UUID id) {
return "Item:" + id;
}
}As Redis provides commands to get and set operations that allow to look for or store multiple objects at once implementation turned out not to be so complicated.
When I started working on it, at first I was afraid that I’ll have to write a lot of low-level code to make it work but luckily with spring’s RedisTemplate it turned out not to be so ugly nor complicated after all. A bit of legacy code allowed me to get to know what spring has to offer in terms of integration with redis, which turned out to be pretty easy to work with ;)
As always sources can be found on my github.
If you've enjoyed or found this post useful you might also like: