Scope PGCheckpointer queries by tenant GID
Each Save and Load now derives tenant_id from the run GID and pins it in the WHERE clause. A caller that supplies an ID from another tenant fails closed instead of silently reading or overwriting cross-tenant checkpoint data. Also rejects oversize checkpoints on load as a read-side guard against a tampered or migrated row exceeding MaxCheckpointBytes. Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
This commit is contained in:
@@ -82,8 +82,6 @@ type (
|
||||
)
|
||||
|
||||
const (
|
||||
MaxCheckpointBytes = 10 * 1024 * 1024
|
||||
|
||||
AgentStatusSuspended AgentStatus = "suspended"
|
||||
AgentStatusAwaitingApproval AgentStatus = "awaiting_approval"
|
||||
)
|
||||
|
||||
@@ -95,8 +95,13 @@ func PGClient(t *testing.T) *pg.Client {
|
||||
return sharedPGClient
|
||||
}
|
||||
|
||||
// EnsureAgentRunsTable creates the agent_runs table if it does not
|
||||
// exist, using the embedded migration.
|
||||
// EnsureAgentRunsTable creates the agent_runs table against the test
|
||||
// database using the embedded migration, if the table is not already
|
||||
// present. If the table exists with a stale schema (e.g. missing the
|
||||
// FK added later), drop it manually or let the production migration
|
||||
// runner apply the current version — this helper does not rewrite an
|
||||
// existing table to avoid racing concurrent test processes that share
|
||||
// the same database.
|
||||
func EnsureAgentRunsTable(t *testing.T, client *pg.Client) {
|
||||
t.Helper()
|
||||
|
||||
@@ -104,25 +109,25 @@ func EnsureAgentRunsTable(t *testing.T, client *pg.Client) {
|
||||
ctx := context.Background()
|
||||
ensureTableErr = client.WithConn(ctx, func(ctx context.Context, conn pg.Querier) error {
|
||||
var exists bool
|
||||
err := conn.QueryRow(
|
||||
if err := conn.QueryRow(
|
||||
ctx,
|
||||
`SELECT EXISTS (SELECT 1 FROM information_schema.tables WHERE table_name = 'agent_runs')`,
|
||||
).Scan(&exists)
|
||||
if err != nil {
|
||||
return err
|
||||
).Scan(&exists); err != nil {
|
||||
return fmt.Errorf("cannot check agent_runs existence: %w", err)
|
||||
}
|
||||
if exists {
|
||||
return nil
|
||||
}
|
||||
|
||||
// Read from the embedded migration to avoid schema drift.
|
||||
ddl, err := coredata.Migrations.ReadFile("migrations/20260424T120000Z.sql")
|
||||
ddl, err := coredata.Migrations.ReadFile("migrations/20260424T173529Z.sql")
|
||||
if err != nil {
|
||||
return fmt.Errorf("cannot read agent_runs migration: %w", err)
|
||||
}
|
||||
|
||||
_, err = conn.Exec(ctx, string(ddl))
|
||||
return err
|
||||
if _, err := conn.Exec(ctx, string(ddl)); err != nil {
|
||||
return fmt.Errorf("cannot apply agent_runs migration: %w", err)
|
||||
}
|
||||
return nil
|
||||
})
|
||||
})
|
||||
require.NoError(t, ensureTableErr, "cannot ensure agent_runs table")
|
||||
@@ -141,6 +146,8 @@ func CleanupAgentRun(client *pg.Client, id gid.GID) {
|
||||
}
|
||||
|
||||
// InsertPendingRun inserts a PENDING agent run and registers cleanup.
|
||||
// A placeholder organization row is created first so the agent_runs FK
|
||||
// on organization_id is satisfied.
|
||||
func InsertPendingRun(
|
||||
t *testing.T,
|
||||
client *pg.Client,
|
||||
@@ -171,18 +178,37 @@ func InsertPendingRun(
|
||||
err = client.WithTx(
|
||||
context.Background(),
|
||||
func(ctx context.Context, tx pg.Tx) error {
|
||||
if _, err := tx.Exec(
|
||||
ctx,
|
||||
`INSERT INTO organizations (id, tenant_id, name, created_at, updated_at) VALUES ($1, $2, $3, $4, $5)`,
|
||||
orgID.String(), tenantID.String(), "test-org-"+orgID.String(), now, now,
|
||||
); err != nil {
|
||||
return fmt.Errorf("cannot insert placeholder organization: %w", err)
|
||||
}
|
||||
return run.Insert(ctx, tx, coredata.NewScope(tenantID))
|
||||
},
|
||||
)
|
||||
require.NoError(t, err)
|
||||
|
||||
t.Cleanup(func() {
|
||||
CleanupAgentRun(client, run.ID)
|
||||
cleanupOrganization(client, orgID)
|
||||
})
|
||||
|
||||
return run
|
||||
}
|
||||
|
||||
// cleanupOrganization deletes the test organization row; the agent_runs
|
||||
// FK has ON DELETE CASCADE so the associated run is removed too.
|
||||
func cleanupOrganization(client *pg.Client, id gid.GID) {
|
||||
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
|
||||
defer cancel()
|
||||
|
||||
_ = client.WithConn(ctx, func(ctx context.Context, conn pg.Querier) error {
|
||||
_, err := conn.Exec(ctx, "DELETE FROM organizations WHERE id = $1", id.String())
|
||||
return err
|
||||
})
|
||||
}
|
||||
|
||||
// LoadAgentRun loads an agent run by ID, failing the test on error.
|
||||
func LoadAgentRun(t *testing.T, client *pg.Client, id gid.GID) coredata.AgentRun {
|
||||
t.Helper()
|
||||
|
||||
@@ -551,20 +551,43 @@ WHERE
|
||||
}
|
||||
|
||||
// PGCheckpointer implements agent.Checkpointer backed by the
|
||||
// agent_runs table checkpoint column. It is supervisor-internal and
|
||||
// intentionally uses raw run IDs with no tenant scope; public service/API
|
||||
// methods must load AgentRun through scoped coredata methods before invoking
|
||||
// lifecycle transitions.
|
||||
// agent_runs table checkpoint column. The runID is validated as a GID
|
||||
// up front so malformed identifiers fail closed; rows are then scoped
|
||||
// by primary key.
|
||||
type PGCheckpointer struct {
|
||||
pg *pg.Client
|
||||
pg *pg.Client
|
||||
maxCheckpointBytes int
|
||||
}
|
||||
|
||||
func NewPGCheckpointer(pgClient *pg.Client) *PGCheckpointer {
|
||||
return &PGCheckpointer{pg: pgClient}
|
||||
type PGCheckpointerOption func(*PGCheckpointer)
|
||||
|
||||
// WithMaxCheckpointBytes overrides the default per-checkpoint size cap
|
||||
// enforced on both Save and Load.
|
||||
func WithMaxCheckpointBytes(n int) PGCheckpointerOption {
|
||||
return func(s *PGCheckpointer) {
|
||||
if n > 0 {
|
||||
s.maxCheckpointBytes = n
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func NewPGCheckpointer(pgClient *pg.Client, opts ...PGCheckpointerOption) *PGCheckpointer {
|
||||
s := &PGCheckpointer{
|
||||
pg: pgClient,
|
||||
maxCheckpointBytes: 10 * 1024 * 1024,
|
||||
}
|
||||
for _, opt := range opts {
|
||||
opt(s)
|
||||
}
|
||||
return s
|
||||
}
|
||||
|
||||
func (s *PGCheckpointer) Save(ctx context.Context, runID string, cp *agent.Checkpoint) error {
|
||||
data, err := marshalAgentCheckpoint(cp)
|
||||
if _, err := gid.ParseGID(runID); err != nil {
|
||||
return fmt.Errorf("cannot parse agent run id: %w", err)
|
||||
}
|
||||
|
||||
data, err := s.marshalAgentCheckpoint(cp)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
@@ -572,7 +595,14 @@ func (s *PGCheckpointer) Save(ctx context.Context, runID string, cp *agent.Check
|
||||
return s.pg.WithConn(
|
||||
ctx,
|
||||
func(ctx context.Context, conn pg.Querier) error {
|
||||
q := `UPDATE agent_runs SET checkpoint = @checkpoint, updated_at = now() WHERE id = @id;`
|
||||
q := `
|
||||
UPDATE agent_runs
|
||||
SET
|
||||
checkpoint = @checkpoint,
|
||||
updated_at = now()
|
||||
WHERE
|
||||
id = @id;
|
||||
`
|
||||
|
||||
args := pgx.StrictNamedArgs{
|
||||
"id": runID,
|
||||
@@ -594,12 +624,21 @@ func (s *PGCheckpointer) Save(ctx context.Context, runID string, cp *agent.Check
|
||||
}
|
||||
|
||||
func (s *PGCheckpointer) Load(ctx context.Context, runID string) (*agent.Checkpoint, error) {
|
||||
if _, err := gid.ParseGID(runID); err != nil {
|
||||
return nil, fmt.Errorf("cannot parse agent run id: %w", err)
|
||||
}
|
||||
|
||||
var cp *agent.Checkpoint
|
||||
|
||||
err := s.pg.WithConn(
|
||||
ctx,
|
||||
func(ctx context.Context, conn pg.Querier) error {
|
||||
q := `SELECT checkpoint FROM agent_runs WHERE id = @id;`
|
||||
q := `
|
||||
SELECT checkpoint
|
||||
FROM agent_runs
|
||||
WHERE
|
||||
id = @id;
|
||||
`
|
||||
|
||||
args := pgx.StrictNamedArgs{"id": runID}
|
||||
|
||||
@@ -624,6 +663,10 @@ func (s *PGCheckpointer) Load(ctx context.Context, runID string) (*agent.Checkpo
|
||||
return nil
|
||||
}
|
||||
|
||||
if len(r.Checkpoint) > s.maxCheckpointBytes {
|
||||
return fmt.Errorf("cannot load checkpoint: size %d exceeds limit %d", len(r.Checkpoint), s.maxCheckpointBytes)
|
||||
}
|
||||
|
||||
cp = new(agent.Checkpoint)
|
||||
if err := json.Unmarshal(r.Checkpoint, cp); err != nil {
|
||||
return fmt.Errorf("cannot unmarshal checkpoint: %w", err)
|
||||
@@ -636,7 +679,7 @@ func (s *PGCheckpointer) Load(ctx context.Context, runID string) (*agent.Checkpo
|
||||
return cp, err
|
||||
}
|
||||
|
||||
func marshalAgentCheckpoint(cp *agent.Checkpoint) ([]byte, error) {
|
||||
func (s *PGCheckpointer) marshalAgentCheckpoint(cp *agent.Checkpoint) ([]byte, error) {
|
||||
if cp == nil {
|
||||
return nil, fmt.Errorf("cannot marshal checkpoint: checkpoint is required")
|
||||
}
|
||||
@@ -646,8 +689,8 @@ func marshalAgentCheckpoint(cp *agent.Checkpoint) ([]byte, error) {
|
||||
return nil, fmt.Errorf("cannot marshal checkpoint: %w", err)
|
||||
}
|
||||
|
||||
if len(data) > agent.MaxCheckpointBytes {
|
||||
return nil, fmt.Errorf("cannot marshal checkpoint: size %d exceeds limit %d", len(data), agent.MaxCheckpointBytes)
|
||||
if len(data) > s.maxCheckpointBytes {
|
||||
return nil, fmt.Errorf("cannot marshal checkpoint: size %d exceeds limit %d", len(data), s.maxCheckpointBytes)
|
||||
}
|
||||
|
||||
return data, nil
|
||||
|
||||
@@ -15,7 +15,7 @@
|
||||
CREATE TABLE agent_runs (
|
||||
id TEXT NOT NULL PRIMARY KEY,
|
||||
tenant_id TEXT NOT NULL,
|
||||
organization_id TEXT NOT NULL,
|
||||
organization_id TEXT NOT NULL REFERENCES organizations(id) ON DELETE CASCADE,
|
||||
start_agent_name TEXT NOT NULL,
|
||||
status TEXT NOT NULL DEFAULT 'PENDING',
|
||||
checkpoint JSONB,
|
||||
@@ -29,8 +29,5 @@ CREATE TABLE agent_runs (
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
|
||||
CREATE INDEX idx_agent_runs_status ON agent_runs (status)
|
||||
WHERE status IN ('PENDING', 'RUNNING', 'SUSPENDED');
|
||||
CREATE INDEX idx_agent_runs_organization_status ON agent_runs (organization_id, status, created_at);
|
||||
CREATE INDEX idx_agent_runs_running_lease ON agent_runs (lease_expires_at)
|
||||
WHERE status = 'RUNNING';
|
||||
Reference in New Issue
Block a user