Add redacted provisioning event history

docs/roadmap.md

@@ -117,7 +117,7 @@
 - [x] начальные строгие JSON schemas v1 для `preflight` и `system.base-install`;
 - [ ] versioned JSON schemas для остальных action type;
 - [ ] restricted root helper через Unix socket;
-- [ ] structured redacted events и progress;
+- [x] structured redacted events, bounded Operator history и progress;
 - [x] `NeedsReconciliation` после истечения execution TTL и неопределённого результата;
 - [ ] desired/factual configuration и drift;
 - [x] запрет параллельных активных заданий на одном Node (безопасный первый вариант).

src/ServerMonitorManager.Control/ControlStore.cs

@@ -10,7 +10,7 @@ namespace ServerMonitorManager.Control;
 
 public sealed partial class ControlStore(IOptions<ControlOptions> options)
 {
-    private const int CurrentSchemaVersion = 4;
+    private const int CurrentSchemaVersion = 5;
     private readonly ControlOptions _options = options.Value;
     private readonly string _connectionString = new SqliteConnectionStringBuilder
     {
@@ -213,6 +213,23 @@ public sealed partial class ControlStore(IOptions<ControlOptions> options)
             await strengthenProvisioningLock.ExecuteNonQueryAsync(cancellationToken);
             await migration.CommitAsync(cancellationToken);
         }
+
+        if (schemaVersion < 5)
+        {
+            await using var migration =
+                (SqliteTransaction)await connection.BeginTransactionAsync(cancellationToken);
+            var addStructuredEventFields = connection.CreateCommand();
+            addStructuredEventFields.Transaction = migration;
+            addStructuredEventFields.CommandText = """
+                ALTER TABLE provisioning_events
+                    ADD COLUMN step TEXT NOT NULL DEFAULT '';
+                ALTER TABLE provisioning_events
+                    ADD COLUMN progress_percent INTEGER NOT NULL DEFAULT 0;
+                PRAGMA user_version = 5;
+                """;
+            await addStructuredEventFields.ExecuteNonQueryAsync(cancellationToken);
+            await migration.CommitAsync(cancellationToken);
+        }
     }
 
     public async Task<ControlMaintenanceResult> MaintainAsync(
@@ -225,9 +242,10 @@ public sealed partial class ControlStore(IOptions<ControlOptions> options)
         var command = connection.CreateCommand();
         command.Transaction = transaction;
         command.CommandText = """
-            INSERT INTO provisioning_events(job_id, recorded_at, event_type, state, message)
+            INSERT INTO provisioning_events(
+                job_id, recorded_at, event_type, state, message, step, progress_percent)
             SELECT id, $now, 'job.expired', 'Cancelled',
-                   'Provisioning job expired before execution.'
+                   'Provisioning job expired before execution.', 'expired', progress_percent
             FROM provisioning_jobs
             WHERE expires_at <= $now
               AND state IN ('Queued', 'AwaitingConfirmation');
@@ -238,9 +256,11 @@ public sealed partial class ControlStore(IOptions<ControlOptions> options)
             WHERE expires_at <= $now
               AND state IN ('Queued', 'AwaitingConfirmation');
             SELECT changes();
-            INSERT INTO provisioning_events(job_id, recorded_at, event_type, state, message)
+            INSERT INTO provisioning_events(
+                job_id, recorded_at, event_type, state, message, step, progress_percent)
             SELECT id, $now, 'job.reconciliation.required', 'NeedsReconciliation',
-                   'Execution lease expired; factual state must be inspected.'
+                   'Execution lease expired; factual state must be inspected.',
+                   'reconcile', progress_percent
             FROM provisioning_jobs
             WHERE expires_at <= $now
               AND state IN ('Preflight', 'Running', 'Verifying');
@@ -251,9 +271,11 @@ public sealed partial class ControlStore(IOptions<ControlOptions> options)
             WHERE expires_at <= $now
               AND state IN ('Preflight', 'Running', 'Verifying');
             SELECT changes();
-            INSERT INTO provisioning_events(job_id, recorded_at, event_type, state, message)
+            INSERT INTO provisioning_events(
+                job_id, recorded_at, event_type, state, message, step, progress_percent)
             SELECT id, $now, 'job.rollback.reconciliation.required', 'NeedsReconciliation',
-                   'Rollback lease expired; factual state must be inspected.'
+                   'Rollback lease expired; factual state must be inspected.',
+                   'rollback-reconcile', progress_percent
             FROM provisioning_jobs
             WHERE expires_at <= $now AND state = 'RollingBack';
             UPDATE provisioning_jobs SET

src/ServerMonitorManager.Control/Program.cs

@@ -484,6 +484,23 @@ control.MapGet("/provisioning/jobs/{id}", async (
     var job = await controlStore.GetProvisioningJobAsync(id, cancellationToken);
     return job is null ? Results.NotFound() : Results.Ok(job);
 });
+control.MapGet("/provisioning/jobs/{id}/events", async (
+    string id,
+    int? limit,
+    ControlStore controlStore,
+    CancellationToken cancellationToken) =>
+{
+    if (!ProvisioningJobValidator.IsValidId(id) || limit is < 1 or > 200)
+    {
+        return Results.ValidationProblem(new Dictionary<string, string[]>
+        {
+            ["provisioningEvents"] = ["Invalid provisioning job id or limit (1-200)."]
+        });
+    }
+    var events = await controlStore.ListProvisioningEventsAsync(
+        id, limit ?? 100, cancellationToken);
+    return events is null ? Results.NotFound() : Results.Ok(events.ToArray());
+});
 control.MapPost("/provisioning/jobs/{id}/confirm", async (
     string id,
     ProvisioningJobCommandRequest request,

src/ServerMonitorManager.Control/ProvisioningStore.cs

@@ -248,7 +248,8 @@ public sealed partial class ControlStore
 
         await WriteProvisioningEventAsync(
             connection, transaction, id, request.EventCode, request.State,
-            request.Message, now, cancellationToken);
+            $"Agent reported '{request.EventCode}' for step '{request.Step}' at "
+            + $"{request.ProgressPercent}%.", now, cancellationToken);
         await WriteIdempotentAsync(
             connection, transaction, operationKey, requestHash, updated,
             SmmJsonContext.Default.ProvisioningJob, cancellationToken);
@@ -266,6 +267,44 @@ public sealed partial class ControlStore
         return updated;
     }
 
+    public async Task<IReadOnlyList<ProvisioningEvent>?> ListProvisioningEventsAsync(
+        string id,
+        int limit,
+        CancellationToken cancellationToken = default)
+    {
+        await using var connection = await OpenAsync(cancellationToken);
+        var exists = connection.CreateCommand();
+        exists.CommandText = "SELECT EXISTS(SELECT 1 FROM provisioning_jobs WHERE id = $id);";
+        exists.Parameters.AddWithValue("$id", id);
+        if (Convert.ToInt32(await exists.ExecuteScalarAsync(cancellationToken)) != 1)
+        {
+            return null;
+        }
+
+        var result = new List<ProvisioningEvent>();
+        var command = connection.CreateCommand();
+        command.CommandText = """
+            SELECT sequence, job_id, recorded_at, event_type, state,
+                   step, progress_percent, message
+            FROM provisioning_events
+            WHERE job_id = $id
+            ORDER BY sequence DESC
+            LIMIT $limit;
+            """;
+        command.Parameters.AddWithValue("$id", id);
+        command.Parameters.AddWithValue("$limit", Math.Clamp(limit, 1, 200));
+        await using var reader = await command.ExecuteReaderAsync(cancellationToken);
+        while (await reader.ReadAsync(cancellationToken))
+        {
+            result.Add(new ProvisioningEvent(
+                reader.GetInt64(0), reader.GetString(1), DateTimeOffset.Parse(reader.GetString(2)),
+                reader.GetString(3), reader.GetString(4), reader.GetString(5),
+                reader.GetInt32(6), reader.GetString(7)));
+        }
+        result.Reverse();
+        return result;
+    }
+
     public async Task<ProvisioningJob?> RetryProvisioningJobAsync(
         string id,
         ProvisioningJobCommandRequest request,
@@ -336,7 +375,7 @@ public sealed partial class ControlStore
 
         await WriteProvisioningEventAsync(
             connection, transaction, id, "job.retry.queued", updated.State,
-            request.Reason, now, cancellationToken);
+            "Operator queued a retry after reconciliation.", now, cancellationToken);
         await WriteIdempotentAsync(
             connection, transaction, operationKey, requestHash, updated,
             SmmJsonContext.Default.ProvisioningJob, cancellationToken);
@@ -418,7 +457,7 @@ public sealed partial class ControlStore
 
         await WriteProvisioningEventAsync(
             connection, transaction, id, "job.rollback.queued", updated.State,
-            request.Reason, now, cancellationToken);
+            "Operator queued a rollback.", now, cancellationToken);
         await WriteIdempotentAsync(
             connection, transaction, operationKey, requestHash, updated,
             SmmJsonContext.Default.ProvisioningJob, cancellationToken);
@@ -494,7 +533,8 @@ public sealed partial class ControlStore
         }
 
         await WriteProvisioningEventAsync(
-            connection, transaction, id, eventType, targetState, request.Reason, now, cancellationToken);
+            connection, transaction, id, eventType, targetState,
+            $"Operator requested '{eventType}'.", now, cancellationToken);
         await WriteIdempotentAsync(
             connection, transaction, operationKey, requestHash, updated,
             SmmJsonContext.Default.ProvisioningJob, cancellationToken);
@@ -569,8 +609,12 @@ public sealed partial class ControlStore
         var command = connection.CreateCommand();
         command.Transaction = transaction;
         command.CommandText = """
-            INSERT INTO provisioning_events(job_id, recorded_at, event_type, state, message)
-            VALUES ($job, $recorded, $event, $state, $message);
+            INSERT INTO provisioning_events(
+                job_id, recorded_at, event_type, state, message, step, progress_percent)
+            SELECT $job, $recorded, $event, $state, $message,
+                   current_step, progress_percent
+            FROM provisioning_jobs
+            WHERE id = $job;
             """;
         command.Parameters.AddWithValue("$job", jobId);
         command.Parameters.AddWithValue("$recorded", recordedAt.ToString("O"));

src/ServerMonitorManager.Core/Contracts.cs

@@ -183,6 +183,16 @@ public sealed record ProvisioningJob(
     string CurrentStep,
     string? LastError);
 
+public sealed record ProvisioningEvent(
+    long Sequence,
+    string JobId,
+    DateTimeOffset RecordedAt,
+    string EventType,
+    string State,
+    string Step,
+    int ProgressPercent,
+    string Message);
+
 public static class ProvisioningJobStates
 {
     public const string Queued = "Queued";

src/ServerMonitorManager.Core/SmmJsonContext.cs

@@ -30,4 +30,6 @@ namespace ServerMonitorManager.Core;
 [JsonSerializable(typeof(ProvisioningJobProgressRequest))]
 [JsonSerializable(typeof(ProvisioningJob))]
 [JsonSerializable(typeof(ProvisioningJob[]))]
+[JsonSerializable(typeof(ProvisioningEvent))]
+[JsonSerializable(typeof(ProvisioningEvent[]))]
 public sealed partial class SmmJsonContext : JsonSerializerContext;

tests/ServerMonitorManager.Control.Tests/ControlApiTests.cs

@@ -116,6 +116,13 @@ public sealed class ControlApiTests : IAsyncDisposable
         Assert.Equal(ServerMonitorManager.Core.ProvisioningJobStates.AwaitingConfirmation, job.State);
         Assert.Equal(HttpStatusCode.OK, (await client.GetAsync(
             $"/api/v1/control/provisioning/jobs/{job.Id}", cancellationToken)).StatusCode);
+        using var eventsResponse = await client.GetAsync(
+            $"/api/v1/control/provisioning/jobs/{job.Id}/events?limit=10", cancellationToken);
+        Assert.Equal(HttpStatusCode.OK, eventsResponse.StatusCode);
+        var events = await eventsResponse.Content.ReadFromJsonAsync<ServerMonitorManager.Core.ProvisioningEvent[]>(
+            cancellationToken);
+        Assert.NotNull(events);
+        Assert.NotEmpty(events);
     }
 
     [Fact]

tests/ServerMonitorManager.Control.Tests/ControlMaintenanceTests.cs

@@ -106,7 +106,7 @@ public sealed class ControlMaintenanceTests : IAsyncDisposable
         await verify.OpenAsync(cancellationToken);
         var version = verify.CreateCommand();
         version.CommandText = "PRAGMA user_version;";
-        Assert.Equal(4L, (long)(await version.ExecuteScalarAsync(cancellationToken))!);
+        Assert.Equal(5L, (long)(await version.ExecuteScalarAsync(cancellationToken))!);
     }
 
     [Fact]

tests/ServerMonitorManager.Control.Tests/ControlStoreTests.cs

@@ -31,7 +31,7 @@ public sealed class ControlStoreTests : IAsyncDisposable
         await migrated.OpenAsync(TestContext.Current.CancellationToken);
         var version = migrated.CreateCommand();
         version.CommandText = "PRAGMA user_version;";
-        Assert.Equal(4L, Convert.ToInt64(await version.ExecuteScalarAsync(TestContext.Current.CancellationToken)));
+        Assert.Equal(5L, Convert.ToInt64(await version.ExecuteScalarAsync(TestContext.Current.CancellationToken)));
         var table = migrated.CreateCommand();
         table.CommandText = "SELECT COUNT(*) FROM pragma_table_info('provisioning_jobs');";
         Assert.Equal(18L, Convert.ToInt64(await table.ExecuteScalarAsync(TestContext.Current.CancellationToken)));
@@ -113,7 +113,7 @@ public sealed class ControlStoreTests : IAsyncDisposable
 
         var preflight = new ProvisioningJobProgressRequest(
             ProvisioningJobStates.Preflight, 10, "inspect-os", "preflight.progress",
-            "Operating system detected.", Guid.NewGuid().ToString());
+            "secret-token-should-never-be-persisted", Guid.NewGuid().ToString());
         Assert.Null(await store.ReportProvisioningProgressAsync(
             "other", created.Id, preflight, cancellationToken));
         var firstProgress = await store.ReportProvisioningProgressAsync(
@@ -170,6 +170,13 @@ public sealed class ControlStoreTests : IAsyncDisposable
         Assert.Equal(ProvisioningJobStates.Running, running!.State);
         Assert.Equal(ProvisioningJobStates.Verifying, verifying!.State);
         Assert.Equal(ProvisioningJobStates.Completed, completed!.State);
+        var events = await store.ListProvisioningEventsAsync(created.Id, 100, cancellationToken);
+        Assert.NotNull(events);
+        var preflightEvent = Assert.Single(events!, item => item.EventType == "preflight.progress");
+        Assert.Equal("inspect-os", preflightEvent.Step);
+        Assert.Equal(10, preflightEvent.ProgressPercent);
+        Assert.DoesNotContain(
+            events!, item => item.Message.Contains("secret-token", StringComparison.Ordinal));
     }
 
     [Fact]