|
| 1 | +"""Add Model Deployment related tables |
| 2 | +
|
| 3 | +Revision ID: 0a3480997c98 |
| 4 | +Revises: 345551145801 |
| 5 | +Create Date: 2025-08-25 10:02:33.684728 |
| 6 | +
|
| 7 | +""" |
| 8 | + |
| 9 | +import sqlalchemy as sa |
| 10 | +from alembic import op |
| 11 | +from sqlalchemy.dialects import postgresql |
| 12 | + |
| 13 | +from ai.backend.common.data.model_deployment.types import ( |
| 14 | + DeploymentStrategy, |
| 15 | + ModelDeploymentStatus, |
| 16 | + ReadinessStatus, |
| 17 | +) |
| 18 | +from ai.backend.common.types import RuntimeVariant, VFolderMount |
| 19 | +from ai.backend.manager.models.base import ( |
| 20 | + GUID, |
| 21 | + DecimalType, |
| 22 | + ResourceSlotColumn, |
| 23 | + StrEnumType, |
| 24 | + StructuredJSONObjectListColumn, |
| 25 | +) |
| 26 | + |
| 27 | +# revision identifiers, used by Alembic. |
| 28 | +revision = "0a3480997c98" |
| 29 | +down_revision = "345551145801" |
| 30 | +branch_labels = None |
| 31 | +depends_on = None |
| 32 | + |
| 33 | + |
| 34 | +def upgrade() -> None: |
| 35 | + # ### commands auto generated by Alembic - please adjust! ### |
| 36 | + op.create_table( |
| 37 | + "blue_green_strategies", |
| 38 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 39 | + sa.Column("pre_switch_wait_time", sa.Integer(), nullable=False), |
| 40 | + sa.Column("switch_timeout", sa.Integer(), nullable=False), |
| 41 | + sa.Column("is_active", sa.Boolean(), nullable=False), |
| 42 | + sa.Column("created_at", sa.DateTime(), nullable=False), |
| 43 | + sa.Column("updated_at", sa.DateTime(), nullable=False), |
| 44 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_blue_green_strategies")), |
| 45 | + ) |
| 46 | + op.create_table( |
| 47 | + "deployment_session_bindings", |
| 48 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 49 | + sa.Column("deployment_id", GUID(), nullable=False), |
| 50 | + sa.Column("session_id", GUID(), nullable=False), |
| 51 | + sa.Column("weight", DecimalType(), nullable=False), |
| 52 | + sa.Column("readiness_status", StrEnumType(ReadinessStatus, length=64), nullable=False), |
| 53 | + sa.Column( |
| 54 | + "created_at", |
| 55 | + sa.DateTime(timezone=True), |
| 56 | + server_default=sa.text("now()"), |
| 57 | + nullable=False, |
| 58 | + ), |
| 59 | + sa.Column( |
| 60 | + "updated_at", |
| 61 | + sa.DateTime(timezone=True), |
| 62 | + server_default=sa.text("now()"), |
| 63 | + nullable=False, |
| 64 | + ), |
| 65 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_deployment_session_bindings")), |
| 66 | + ) |
| 67 | + op.create_table( |
| 68 | + "model_deployment_auto_scaling_rules", |
| 69 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 70 | + sa.Column("model_deployment_id", GUID(), nullable=False), |
| 71 | + sa.Column("metric_source", sa.String(), nullable=False), |
| 72 | + sa.Column("metric_name", sa.String(), nullable=False), |
| 73 | + sa.Column("min_threshold", DecimalType(), nullable=True), |
| 74 | + sa.Column("max_threshold", DecimalType(), nullable=True), |
| 75 | + sa.Column("step_size", sa.Integer(), nullable=False), |
| 76 | + sa.Column("time_window", sa.Integer(), nullable=False), |
| 77 | + sa.Column("min_replicas", sa.Integer(), nullable=True), |
| 78 | + sa.Column("max_replicas", sa.Integer(), nullable=True), |
| 79 | + sa.Column( |
| 80 | + "created_at", |
| 81 | + sa.DateTime(timezone=True), |
| 82 | + server_default=sa.text("now()"), |
| 83 | + nullable=False, |
| 84 | + ), |
| 85 | + sa.Column("last_triggered_at", sa.DateTime(timezone=True), nullable=False), |
| 86 | + sa.CheckConstraint( |
| 87 | + "(min_threshold IS NOT NULL AND max_threshold IS NULL) OR (min_threshold IS NULL AND max_threshold IS NOT NULL)", |
| 88 | + name=op.f("ck_model_deployment_auto_scaling_rules_check_single_threshold"), |
| 89 | + ), |
| 90 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployment_auto_scaling_rules")), |
| 91 | + sa.UniqueConstraint( |
| 92 | + "model_deployment_id", "metric_name", name="uq_model_deployment_metric" |
| 93 | + ), |
| 94 | + ) |
| 95 | + op.create_table( |
| 96 | + "model_deployment_tokens", |
| 97 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 98 | + sa.Column("token", sa.String(), nullable=False), |
| 99 | + sa.Column("model_deployment_id", GUID(), nullable=False), |
| 100 | + sa.Column( |
| 101 | + "created_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True |
| 102 | + ), |
| 103 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployment_tokens")), |
| 104 | + ) |
| 105 | + op.create_table( |
| 106 | + "model_deployments", |
| 107 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 108 | + sa.Column("name", sa.String(), nullable=False), |
| 109 | + sa.Column("status", StrEnumType(ModelDeploymentStatus, length=64), nullable=False), |
| 110 | + sa.Column("tags", sa.String(), nullable=False), |
| 111 | + sa.Column("endpoint_url", sa.String(), nullable=True), |
| 112 | + sa.Column("preferred_domain_name", sa.String(), nullable=True), |
| 113 | + sa.Column("open_to_public", sa.Boolean(), nullable=False), |
| 114 | + sa.Column("desired_replica_count", sa.Integer(), nullable=False), |
| 115 | + sa.Column("created_user_id", GUID(), nullable=False), |
| 116 | + sa.Column("current_revision_id", GUID(), nullable=False), |
| 117 | + sa.Column( |
| 118 | + "deployment_strategy_type", StrEnumType(DeploymentStrategy, length=64), nullable=False |
| 119 | + ), |
| 120 | + sa.Column("deployment_strategy_id", GUID(), nullable=False), |
| 121 | + sa.Column( |
| 122 | + "created_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True |
| 123 | + ), |
| 124 | + sa.Column( |
| 125 | + "updated_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True |
| 126 | + ), |
| 127 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployments")), |
| 128 | + ) |
| 129 | + op.create_index( |
| 130 | + op.f("ix_model_deployments_created_user_id"), |
| 131 | + "model_deployments", |
| 132 | + ["created_user_id"], |
| 133 | + unique=False, |
| 134 | + ) |
| 135 | + op.create_index(op.f("ix_model_deployments_name"), "model_deployments", ["name"], unique=False) |
| 136 | + op.create_table( |
| 137 | + "model_revisions", |
| 138 | + sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False), |
| 139 | + sa.Column("name", sa.String(), nullable=False), |
| 140 | + sa.Column("model_deployment_id", GUID(), nullable=False), |
| 141 | + sa.Column("tags", sa.TEXT(), nullable=False), |
| 142 | + sa.Column("runtime_variant", StrEnumType(RuntimeVariant, length=64), nullable=False), |
| 143 | + sa.Column( |
| 144 | + "inference_runtime_config", postgresql.JSONB(astext_type=sa.Text()), nullable=False |
| 145 | + ), |
| 146 | + sa.Column("environment_variables", postgresql.JSONB(astext_type=sa.Text()), nullable=False), |
| 147 | + sa.Column("model_vfolder_id", GUID(), nullable=False), |
| 148 | + sa.Column("model_mount_destination", sa.String(), server_default="/models", nullable=False), |
| 149 | + sa.Column("model_definition_path", sa.String(), nullable=False), |
| 150 | + sa.Column("image_id", GUID(), nullable=False), |
| 151 | + sa.Column("cluster_mode", sa.String(), server_default="single-node", nullable=False), |
| 152 | + sa.Column("cluster_size", sa.Integer(), server_default="1", nullable=False), |
| 153 | + sa.Column("resource_group", sa.String(), nullable=False), |
| 154 | + sa.Column("resource_slots", ResourceSlotColumn(), nullable=False), |
| 155 | + sa.Column("resource_opts", postgresql.JSONB(astext_type=sa.Text()), nullable=False), |
| 156 | + sa.Column( |
| 157 | + "extra_mount", |
| 158 | + StructuredJSONObjectListColumn(VFolderMount), |
| 159 | + server_default="[]", |
| 160 | + nullable=False, |
| 161 | + ), |
| 162 | + sa.Column( |
| 163 | + "created_at", |
| 164 | + sa.DateTime(timezone=True), |
| 165 | + server_default=sa.text("now()"), |
| 166 | + nullable=False, |
| 167 | + ), |
| 168 | + sa.PrimaryKeyConstraint("id", name=op.f("pk_model_revisions")), |
| 169 | + ) |
| 170 | + # ### end Alembic commands ### |
| 171 | + |
| 172 | + |
| 173 | +def downgrade() -> None: |
| 174 | + # ### commands auto generated by Alembic - please adjust! ### |
| 175 | + op.drop_table("model_revisions") |
| 176 | + op.drop_index(op.f("ix_model_deployments_name"), table_name="model_deployments") |
| 177 | + op.drop_index(op.f("ix_model_deployments_created_user_id"), table_name="model_deployments") |
| 178 | + op.drop_table("model_deployments") |
| 179 | + op.drop_table("model_deployment_tokens") |
| 180 | + op.drop_table("model_deployment_auto_scaling_rules") |
| 181 | + op.drop_table("deployment_session_bindings") |
| 182 | + op.drop_table("blue_green_strategies") |
| 183 | + # ### end Alembic commands ### |
0 commit comments