Skip to content

unify caching across workers to maximize cache hits #977

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jfallows merged 15 commits into from
May 10, 2024
Merged

unify caching across workers to maximize cache hits #977

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
d6dffb6
unify caching across workers to maximize cache hits
ankitk-me Apr 24, 2024
d778b9e
addressing review comments
ankitk-me Apr 25, 2024
6844810
addressing review comments
ankitk-me Apr 29, 2024
04ee960
including event log for FutureCompletedExceptionally & StaleSchemaServed
ankitk-me Apr 29, 2024
da28dd7
update to handle unifying cache getting access by other catalog instance
ankitk-me Apr 30, 2024
4602867
checkstyle fix
ankitk-me Apr 30, 2024
e6d7cf5
using computeIfAbsent
ankitk-me Apr 30, 2024
d92bc16
addressing review comments
ankitk-me May 2, 2024
0027f5f
adding support for progressive back off
ankitk-me May 3, 2024
3485813
addressing review comments
ankitk-me May 7, 2024
dc244a6
stale schema logging fix
ankitk-me May 7, 2024
0ce3f55
addressing review comments
ankitk-me May 8, 2024
62c8220
updating IT to use zilla.yaml instead of mock & handler
ankitk-me May 9, 2024
4fba34f
Merge remote-tracking branch 'origin' into unify_caching
ankitk-me May 9, 2024
0fd0060
addressing review comments
ankitk-me May 10, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion ...main/java/io/aklivity/zilla/runtime/catalog/karapace/internal/KarapaceCatalogContext.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,10 @@
*/
package io.aklivity.zilla.runtime.catalog.karapace.internal;

import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;

import io.aklivity.zilla.runtime.catalog.karapace.internal.config.KarapaceOptionsConfig;
import io.aklivity.zilla.runtime.engine.EngineContext;
import io.aklivity.zilla.runtime.engine.catalog.CatalogContext;
Expand All @@ -23,17 +27,22 @@
public class KarapaceCatalogContext implements CatalogContext
{
private final EngineContext context;
private final ConcurrentMap<Integer, CompletableFuture<String>> cachedSchemas;
private final ConcurrentMap<Integer, CompletableFuture<CachedSchemaId>> cachedSchemaIds;

public KarapaceCatalogContext(
EngineContext context)
{
this.context = context;
this.cachedSchemas = new ConcurrentHashMap<>();
this.cachedSchemaIds = new ConcurrentHashMap<>();
}

@Override
public CatalogHandler attach(
CatalogConfig catalog)
{
return new KarapaceCatalogHandler(KarapaceOptionsConfig.class.cast(catalog.options), context, catalog.id);
return new KarapaceCatalogHandler(KarapaceOptionsConfig.class.cast(catalog.options), context, catalog.id,
cachedSchemas, cachedSchemaIds);
}
}
126 changes: 112 additions & 14 deletions ...main/java/io/aklivity/zilla/runtime/catalog/karapace/internal/KarapaceCatalogHandler.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,10 @@
import java.net.http.HttpResponse;
import java.nio.ByteOrder;
import java.text.MessageFormat;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.ExecutionException;
import java.util.zip.CRC32C;

import org.agrona.BitUtil;
Expand Down Expand Up @@ -54,11 +58,23 @@ public class KarapaceCatalogHandler implements CatalogHandler
private final long maxAgeMillis;
private final KarapaceEventContext event;
private final long catalogId;
private final ConcurrentMap<Integer, CompletableFuture<String>> cachedSchemas;
private final ConcurrentMap<Integer, CompletableFuture<CachedSchemaId>> cachedSchemaIds;

public KarapaceCatalogHandler(
KarapaceOptionsConfig config,
EngineContext context,
long catalogId)
{
this(config, context, catalogId, new ConcurrentHashMap<>(), new ConcurrentHashMap<>());
}

public KarapaceCatalogHandler(
KarapaceOptionsConfig config,
EngineContext context,
long catalogId,
ConcurrentMap<Integer, CompletableFuture<String>> cachedSchemas,
ConcurrentMap<Integer, CompletableFuture<CachedSchemaId>> cachedSchemaIds)
{
this.baseUrl = config.url;
this.client = HttpClient.newHttpClient();
Expand All @@ -69,24 +85,48 @@ public KarapaceCatalogHandler(
this.maxAgeMillis = config.maxAge.toMillis();
this.event = new KarapaceEventContext(context);
this.catalogId = catalogId;
this.cachedSchemas = cachedSchemas;
this.cachedSchemaIds = cachedSchemaIds;
}

@Override
public String resolve(
int schemaId)
{
String schema;
String schema = null;
if (schemas.containsKey(schemaId))
{
schema = schemas.get(schemaId);
}
else
{
String response = sendHttpRequest(MessageFormat.format(SCHEMA_PATH, schemaId));
schema = response != null ? request.resolveSchemaResponse(response) : null;
if (schema != null)
CompletableFuture<String> future = cachedSchemas.get(schemaId);
if (future == null)
{
CompletableFuture<String> newFuture = new CompletableFuture<String>();
future = cachedSchemas.putIfAbsent(schemaId, newFuture);
if (future == null)
{
newFuture = CompletableFuture.supplyAsync(() ->
{
String response = sendHttpRequest(MessageFormat.format(SCHEMA_PATH, schemaId));
return response != null ? request.resolveSchemaResponse(response) : null;
});
future = newFuture;

@jfallows jfallows Apr 25, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think we can reassign newFuture here, as the value being lost is the one that already made it into the concurrent map, so if we reassign it then the one in the map would not be completed, agree?

}
}
assert future != null;
try
{
schema = future.get();
if (schema != null)
{
schemas.put(schemaId, schema);
}
}
catch (ExecutionException | InterruptedException e)
{
schemas.put(schemaId, schema);
// TODO: log an event
}
}
return schema;
Expand All @@ -97,22 +137,80 @@ public int resolve(
String subject,
String version)
{
int schemaId;
int schemaId = NO_SCHEMA_ID;

int checkSum = generateCRC32C(subject, version);
if (schemaIds.containsKey(checkSum) &&
(System.currentTimeMillis() - schemaIds.get(checkSum).timestamp) < maxAgeMillis)
int schemaKey = generateCRC32C(subject, version);
if (schemaIds.containsKey(schemaKey) &&
(System.currentTimeMillis() - schemaIds.get(schemaKey).timestamp) < maxAgeMillis)
{
schemaId = schemaIds.get(checkSum).id;
schemaId = schemaIds.get(schemaKey).id;
}
else
{
String response = sendHttpRequest(MessageFormat.format(SUBJECT_VERSION_PATH, subject, version));
schemaId = response != null ? request.resolveResponse(response) : NO_SCHEMA_ID;
if (schemaId != NO_SCHEMA_ID)
CompletableFuture<CachedSchemaId> future = cachedSchemaIds.get(schemaKey);
if (future == null)
{
schemaIds.put(checkSum, new CachedSchemaId(System.currentTimeMillis(), schemaId));
CompletableFuture<CachedSchemaId> newFuture = new CompletableFuture<>();
future = cachedSchemaIds.putIfAbsent(schemaKey, newFuture);
if (future == null)
{
newFuture = CompletableFuture.supplyAsync(() ->
{
String response = sendHttpRequest(MessageFormat.format(SUBJECT_VERSION_PATH, subject, version));
return new CachedSchemaId(System.currentTimeMillis(),
response != null ? request.resolveResponse(response) : NO_SCHEMA_ID);
});
future = newFuture;

@jfallows jfallows Apr 25, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same issue here as above, agree?

}
}
assert future != null;
try
{
CachedSchemaId cachedSchemaId = future.get();
schemaId = cachedSchemaId.id;
if (schemaId != NO_SCHEMA_ID && (System.currentTimeMillis() - cachedSchemaId.timestamp) < maxAgeMillis)
{
schemaIds.put(schemaKey, cachedSchemaId);
}
else
{
CompletableFuture<CachedSchemaId> newFuture =
cachedSchemaIds.computeIfPresent(schemaKey, (key, existingFuture) ->
{
if (existingFuture.isDone() || existingFuture.isCompletedExceptionally())
{
CompletableFuture<CachedSchemaId> id = CompletableFuture.supplyAsync(() ->
{
String response = sendHttpRequest(MessageFormat.format(SUBJECT_VERSION_PATH,
subject, version));
return new CachedSchemaId(System.currentTimeMillis(),
response != null ? request.resolveResponse(response) : NO_SCHEMA_ID);
});
return id;
}
else
{
return existingFuture;
}
});

@jfallows jfallows Apr 25, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we can perhaps simplify the logic by creating the candidate newFuture and calling:

CompletableFuture<> newFuture = new CompletableFuture<CachedSchemaId>();
CompletableFuture<> future = cachedSchemaIds.merge(schemaKey, newFuture, (v1, v2) -> v1 != null && !v1.isDone() ? v1 : v2);
if (future == newFuture)
{
    // send request
    // complete future
}
cachedSchemaId  = future.get();

cachedSchemaId = newFuture.get();
schemaId = cachedSchemaId.id;
if (schemaId != NO_SCHEMA_ID)
{
schemaIds.put(schemaKey, cachedSchemaId);
}
}
}
Comment on lines +139 to +270

@jfallows jfallows Apr 24, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this needed to handle the case where remote access fails, but we can serve from (stale) cache instead?

@ankitk-me ankitk-me Apr 25, 2024

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yes.

catch (ExecutionException | InterruptedException e)
{
// TODO: log an event
}
}

if (schemaId == NO_SCHEMA_ID && schemaIds.containsKey(schemaKey))
{
schemaId = schemaIds.get(schemaKey).id;
// TODO: log an event to notify, that stale schemaId was returned
}
return schemaId;
}
Expand Down