fix smoother migrations

api/v1alpha1/hyperbytedbcluster_types.go

@@ -229,10 +229,21 @@ type ChDBSpec struct {
 	// +optional
 	SessionDataPath string `json:"sessionDataPath,omitempty"`
 
-	// Ignored by hyperbytedb (libchdb is a process-global singleton). Retained
-	// for API stability; always written as 1 in config.toml.
-	// +kubebuilder:default=1
+	// Sets both query and write pool sizes when QueryPoolSize and WritePoolSize
+	// are unset. Defaults to 1 when all pool fields are unset.
+	// +optional
+	// +kubebuilder:validation:Minimum=1
 	PoolSize int32 `json:"poolSize,omitempty"`
+
+	// chDB connections reserved for queries. Isolated from ingest/flush.
+	// +optional
+	// +kubebuilder:validation:Minimum=1
+	QueryPoolSize int32 `json:"queryPoolSize,omitempty"`
+
+	// chDB connections reserved for ingest and flush (Arrow WAL build, INSERTs).
+	// +optional
+	// +kubebuilder:validation:Minimum=1
+	WritePoolSize int32 `json:"writePoolSize,omitempty"`
 }
 
 type AuthSpec struct {
@@ -264,7 +275,7 @@ type ClusterTuningSpec struct {
 	// +kubebuilder:default=5
 	ReplicationMaxRetries int32 `json:"replicationMaxRetries,omitempty"`
 
-	// +kubebuilder:default=300
+	// +kubebuilder:default=1000
 	RaftHeartbeatIntervalMs int32 `json:"raftHeartbeatIntervalMs,omitempty"`
 
 	// +kubebuilder:default=1000
@@ -306,6 +317,12 @@ type ClusterTuningSpec struct {
 	// TLS for inter-node replication traffic.
 	// +optional
 	TLS *TLSSpec `json:"tls,omitempty"`
+
+	// Seconds to wait after excluding a backend from the proxy before
+	// deleting its pod. Gives in-flight requests time to drain.
+	// +kubebuilder:default=10
+	// +optional
+	DrainWaitSecs int32 `json:"drainWaitSecs,omitempty"`
 }
 
 // ReplicationSpec controls coordinator-side replication (how this node's
@@ -543,6 +560,40 @@ const (
 	ClusterPhaseFailed       ClusterPhase = "Failed"
 )
 
+// RollingRestartPhase tracks which step of the proxy-coordinated rolling
+// restart the operator is currently executing.
+type RollingRestartPhase string
+
+const (
+	RollingRestartExcluding    RollingRestartPhase = "Excluding"
+	RollingRestartDraining     RollingRestartPhase = "Draining"
+	RollingRestartWaitingReady RollingRestartPhase = "WaitingReady"
+	RollingRestartIncluding    RollingRestartPhase = "Including"
+	RollingRestartCompleted    RollingRestartPhase = "Completed"
+)
+
+// RollingRestartState tracks pod-by-pod proxy exclusion during rolling upgrades.
+// Nil when no rolling restart is in progress.
+type RollingRestartState struct {
+	// Ordinal of the pod currently being restarted.
+	CurrentOrdinal int32 `json:"currentOrdinal"`
+
+	// Total number of pod ordinals to cycle through.
+	TotalOrdinals int32 `json:"totalOrdinals"`
+
+	// Current phase of the restart state machine.
+	Phase RollingRestartPhase `json:"phase"`
+
+	// When the current phase started (used to compute drain wait).
+	PhaseStartedAt metav1.Time `json:"phaseStartedAt"`
+
+	// IP of the pod being excluded (set during Excluding phase).
+	OldPodIP string `json:"oldPodIP,omitempty"`
+
+	// True once the proxy confirms the backend is excluded.
+	ExcludeConfirmed bool `json:"excludeConfirmed"`
+}
+
 // MemberStatus describes the observed state of a single cluster member.
 type MemberStatus struct {
 	// Stable identifier derived from the StatefulSet ordinal.
@@ -604,6 +655,11 @@ type HyperbytedbClusterStatus struct {
 	// +optional
 	ConfigHash string `json:"configHash,omitempty"`
 
+	// Tracks pod-by-pod proxy exclusion during rolling upgrades.
+	// Nil when no rolling restart is in progress.
+	// +optional
+	RollingRestart *RollingRestartState `json:"rollingRestart,omitempty"`
+
 	// +listType=map
 	// +listMapKey=type
 	// +optional

config/crd/bases/hyperbytedb.hyperbyte.cloud_hyperbytedbclusters.yaml

@@ -2977,20 +2977,39 @@ spec:
               chdb:
                 properties:
                   poolSize:
-                    default: 1
                     description: |-
-                      Ignored by hyperbytedb (libchdb is a process-global singleton). Retained
-                      for API stability; always written as 1 in config.toml.
+                      Sets both query and write pool sizes when QueryPoolSize and WritePoolSize
+                      are unset. Defaults to 1 when all pool fields are unset.
                     format: int32
+                    minimum: 1
+                    type: integer
+                  queryPoolSize:
+                    description: chDB connections reserved for queries. Isolated from
+                      ingest/flush.
+                    format: int32
+                    minimum: 1
                     type: integer
                   sessionDataPath:
                     description: |-
                       chDB session directory inside the data volume. Defaults to
                       /var/lib/hyperbytedb/chdb when unset.
                     type: string
+                  writePoolSize:
+                    description: chDB connections reserved for ingest and flush (Arrow
+                      WAL build, INSERTs).
+                    format: int32
+                    minimum: 1
+                    type: integer
                 type: object
               cluster:
                 properties:
+                  drainWaitSecs:
+                    default: 10
+                    description: |-
+                      Seconds to wait after excluding a backend from the proxy before
+                      deleting its pod. Gives in-flight requests time to drain.
+                    format: int32
+                    type: integer
                   heartbeatIntervalSecs:
                     default: 2
                     format: int32
@@ -3004,7 +3023,7 @@ spec:
                     format: int32
                     type: integer
                   raftHeartbeatIntervalMs:
-                    default: 300
+                    default: 1000
                     format: int32
                     type: integer
                   raftSnapshotThreshold:
@@ -3977,6 +3996,41 @@ spec:
                 description: 'Replication convergence state: Healthy, Lagging, Diverged,
                   Unknown.'
                 type: string
+              rollingRestart:
+                description: |-
+                  Tracks pod-by-pod proxy exclusion during rolling upgrades.
+                  Nil when no rolling restart is in progress.
+                properties:
+                  currentOrdinal:
+                    description: Ordinal of the pod currently being restarted.
+                    format: int32
+                    type: integer
+                  excludeConfirmed:
+                    description: True once the proxy confirms the backend is excluded.
+                    type: boolean
+                  oldPodIP:
+                    description: IP of the pod being excluded (set during Excluding
+                      phase).
+                    type: string
+                  phase:
+                    description: Current phase of the restart state machine.
+                    type: string
+                  phaseStartedAt:
+                    description: When the current phase started (used to compute drain
+                      wait).
+                    format: date-time
+                    type: string
+                  totalOrdinals:
+                    description: Total number of pod ordinals to cycle through.
+                    format: int32
+                    type: integer
+                required:
+                - currentOrdinal
+                - excludeConfirmed
+                - phase
+                - phaseStartedAt
+                - totalOrdinals
+                type: object
             type: object
         required:
         - spec