feat: add tolerance configuration

diennea · Nov 13, 2024 · f2d5f4a · f2d5f4a
1 parent bbdb299
commit f2d5f4a
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 1 deletion.
diff --git a/carapace-server/src/main/java/org/carapaceproxy/core/RuntimeServerConfiguration.java b/carapace-server/src/main/java/org/carapaceproxy/core/RuntimeServerConfiguration.java
@@ -120,6 +120,7 @@ public class RuntimeServerConfiguration {
     private int healthProbePeriod = DEFAULT_PROBE_PERIOD;
     private int healthConnectTimeout = 5_000;
     private long warmupPeriod = DEFAULT_WARMUP_PERIOD;
+    private boolean tolerant = false;
     private int dynamicCertificatesManagerPeriod = 0;
     private int keyPairsSize = DEFAULT_KEYPAIRS_SIZE;
     private Set<String> domainsCheckerIPAddresses;
@@ -238,6 +239,9 @@ public void configure(ConfigurationStore properties) throws ConfigurationNotVali
         warmupPeriod = properties.getLong("healthmanager.warmupperiod", DEFAULT_WARMUP_PERIOD);
         LOG.log(Level.INFO, "healthmanager.warmupperiod={0}", warmupPeriod);
 
+        tolerant = properties.getBoolean("healthmanager.tolerant", false);
+        LOG.log(Level.INFO, "healthmanager.tolerant={0}", tolerant);
+
         healthConnectTimeout = properties.getInt("healthmanager.connecttimeout", healthConnectTimeout);
         LOG.log(Level.INFO, "healthmanager.connecttimeout={0}", healthConnectTimeout);
         if (healthConnectTimeout < 0) {

diff --git a/carapace-server/src/main/java/org/carapaceproxy/server/backends/BackendHealthManager.java b/carapace-server/src/main/java/org/carapaceproxy/server/backends/BackendHealthManager.java
@@ -62,11 +62,13 @@ public class BackendHealthManager implements Runnable {
     // keep track of start() calling
     private volatile boolean started;
     private volatile long warmupPeriod;
+    private volatile boolean tolerant;
 
     public BackendHealthManager(final RuntimeServerConfiguration conf, final EndpointMapper mapper) {
         this.mapper = mapper;
         this.connectTimeout = conf.getHealthConnectTimeout();
         this.warmupPeriod = conf.getWarmupPeriod();
+        this.tolerant = conf.isTolerant();
 
         // will be overridden before start
         this.period = DEFAULT_PERIOD;
@@ -131,6 +133,11 @@ public synchronized void reloadConfiguration(RuntimeServerConfiguration newConfi
             LOG.info("Applying new warmup period of {} ms", this.warmupPeriod);
         }
 
+        if (this.tolerant != newConfiguration.isTolerant()) {
+            this.tolerant = newConfiguration.isTolerant();
+            LOG.info("Applying new health tolerance configuration {}; cold backends now {} exceed safe capacity", this.tolerant, this.tolerant ? "may" : "may not");
+        }
+
         this.mapper = mapper;
 
         if (restart || started) {
@@ -210,7 +217,7 @@ public boolean exceedsCapacity(final String backendId) {
             return false;
         }
         final BackendHealthStatus backendStatus = getBackendStatus(backendConfiguration.hostPort());
-        return backendConfiguration.safeCapacity() > backendStatus.getConnections();
+        return backendConfiguration.safeCapacity() > backendStatus.getConnections() && !this.tolerant;
     }
 
     @VisibleForTesting