Allow exporting via signed URL

This patch deprecates the pprof report type for a more generic export type, which both extends the functionality to allow for more formats to be possible in the future as well as allow either serving the download content inline in the response like the pprof report does, or if Parca is configured this way, then upload the export to object storage and serve a download URL that is a signed URL from object storage.
parca-dev · May 24, 2024 · 6822f40 · 6822f40
1 parent 9c82f75
commit 6822f40
Show file tree

Hide file tree

Showing 11 changed files with 2,138 additions and 617 deletions.
diff --git a/gen/proto/go/parca/query/v1alpha1/query.pb.go b/gen/proto/go/parca/query/v1alpha1/query.pb.go
diff --git a/gen/proto/go/parca/query/v1alpha1/query_vtproto.pb.go b/gen/proto/go/parca/query/v1alpha1/query_vtproto.pb.go
diff --git a/gen/proto/swagger/parca/query/v1alpha1/query.swagger.json b/gen/proto/swagger/parca/query/v1alpha1/query.swagger.json
@@ -299,7 +299,7 @@
           },
           {
             "name": "reportType",
-            "description": "report_type is the type of report to return\n\n - REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels",
+            "description": "report_type is the type of report to return\n\n - REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels\n - REPORT_TYPE_EXPORT: REPORT_TYPE_EXPORT contains the profile as a downloadable file",
             "in": "query",
             "required": false,
             "type": "string",
@@ -312,7 +312,8 @@
               "REPORT_TYPE_FLAMEGRAPH_ARROW",
               "REPORT_TYPE_SOURCE",
               "REPORT_TYPE_TABLE_ARROW",
-              "REPORT_TYPE_PROFILE_METADATA"
+              "REPORT_TYPE_PROFILE_METADATA",
+              "REPORT_TYPE_EXPORT"
             ],
             "default": "REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED"
           },
@@ -390,6 +391,18 @@
             "in": "query",
             "required": false,
             "type": "boolean"
+          },
+          {
+            "name": "exportFormat",
+            "description": "the format to respond the data in for the export\n\n - FORMAT_UNSPECIFIED: FORMAT_UNSPECIFIED unspecified\n - FORMAT_PPROF: DOWNLOAD_FORMAT_PPROF pprof format",
+            "in": "query",
+            "required": false,
+            "type": "string",
+            "enum": [
+              "FORMAT_UNSPECIFIED",
+              "FORMAT_PPROF"
+            ],
+            "default": "FORMAT_UNSPECIFIED"
           }
         ],
         "tags": [
@@ -581,10 +594,11 @@
         "REPORT_TYPE_FLAMEGRAPH_ARROW",
         "REPORT_TYPE_SOURCE",
         "REPORT_TYPE_TABLE_ARROW",
-        "REPORT_TYPE_PROFILE_METADATA"
+        "REPORT_TYPE_PROFILE_METADATA",
+        "REPORT_TYPE_EXPORT"
       ],
       "default": "REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED",
-      "description": "- REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels",
+      "description": "- REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels\n - REPORT_TYPE_EXPORT: REPORT_TYPE_EXPORT contains the profile as a downloadable file",
       "title": "ReportType is the type of report to return"
     },
     "metastorev1alpha1Function": {
@@ -813,6 +827,10 @@
             "$ref": "#/definitions/v1alpha1Filter"
           },
           "title": "a set of filter to apply to the query request"
+        },
+        "exportFormat": {
+          "$ref": "#/definitions/v1alpha1Format",
+          "title": "the format to respond the data in for the export"
         }
       },
       "title": "QueryRequest is a request for a profile query"
@@ -853,6 +871,10 @@
           "$ref": "#/definitions/v1alpha1ProfileMetadata",
           "title": "profile_metadata contains metadata about the profile i.e. binaries, labels"
         },
+        "export": {
+          "$ref": "#/definitions/v1alpha1Export",
+          "description": "export is the response for an export request."
+        },
         "total": {
           "type": "string",
           "format": "int64",
@@ -1068,6 +1090,24 @@
       },
       "title": "DiffProfile contains parameters for a profile diff request"
     },
+    "v1alpha1Export": {
+      "type": "object",
+      "properties": {
+        "format": {
+          "$ref": "#/definitions/v1alpha1Format",
+          "description": "The format of the download."
+        },
+        "inline": {
+          "$ref": "#/definitions/v1alpha1InlineExport",
+          "description": "InlineExport is the response for an export request where actual content\nis inlined in the response."
+        },
+        "url": {
+          "$ref": "#/definitions/v1alpha1URLExport",
+          "description": "URL is the response for an export request where actual content\nis to be downloaded from the provided URL."
+        }
+      },
+      "description": "Export is the response for an export request."
+    },
     "v1alpha1Filter": {
       "type": "object",
       "properties": {
@@ -1256,6 +1296,16 @@
       },
       "title": "FlamegraphRootNode is a root node of a flame graph"
     },
+    "v1alpha1Format": {
+      "type": "string",
+      "enum": [
+        "FORMAT_UNSPECIFIED",
+        "FORMAT_PPROF"
+      ],
+      "default": "FORMAT_UNSPECIFIED",
+      "description": "- FORMAT_UNSPECIFIED: FORMAT_UNSPECIFIED unspecified\n - FORMAT_PPROF: DOWNLOAD_FORMAT_PPROF pprof format",
+      "title": "format is the format to offer the data in for the download"
+    },
     "v1alpha1FrameFilter": {
       "type": "object",
       "properties": {
@@ -1289,6 +1339,17 @@
       },
       "title": "GroupBy encapsulates the repeated fields to group by"
     },
+    "v1alpha1InlineExport": {
+      "type": "object",
+      "properties": {
+        "content": {
+          "type": "string",
+          "format": "byte",
+          "description": "Content of the export."
+        }
+      },
+      "description": "InlineExport is the response for an export request where actual content is\ninlined in the response."
+    },
     "v1alpha1LabelSet": {
       "type": "object",
       "properties": {
@@ -1677,6 +1738,16 @@
       },
       "title": "TopNodeMeta is the metadata for a given node"
     },
+    "v1alpha1URLExport": {
+      "type": "object",
+      "properties": {
+        "url": {
+          "type": "string",
+          "description": "The URL to download the content from."
+        }
+      },
+      "description": "URLExport is the response for an export request where actual content is\nto be downloaded from the provided URL."
+    },
     "v1alpha1ValuesResponse": {
       "type": "object",
       "properties": {

diff --git a/gen/proto/swagger/parca/share/v1alpha1/share.swagger.json b/gen/proto/swagger/parca/share/v1alpha1/share.swagger.json
@@ -28,10 +28,11 @@
         "REPORT_TYPE_FLAMEGRAPH_ARROW",
         "REPORT_TYPE_SOURCE",
         "REPORT_TYPE_TABLE_ARROW",
-        "REPORT_TYPE_PROFILE_METADATA"
+        "REPORT_TYPE_PROFILE_METADATA",
+        "REPORT_TYPE_EXPORT"
       ],
       "default": "REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED",
-      "description": "- REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels",
+      "description": "- REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED: REPORT_TYPE_FLAMEGRAPH_UNSPECIFIED unspecified\n - REPORT_TYPE_PPROF: REPORT_TYPE_PPROF unspecified\n - REPORT_TYPE_TOP: REPORT_TYPE_TOP unspecified\n - REPORT_TYPE_CALLGRAPH: REPORT_TYPE_CALLGRAPH unspecified\n - REPORT_TYPE_FLAMEGRAPH_TABLE: REPORT_TYPE_FLAMEGRAPH_TABLE unspecified\n - REPORT_TYPE_FLAMEGRAPH_ARROW: REPORT_TYPE_FLAMEGRAPH_ARROW unspecified\n - REPORT_TYPE_SOURCE: REPORT_TYPE_SOURCE contains source code annotated with profiling information\n - REPORT_TYPE_TABLE_ARROW: REPORT_TYPE_TABLE_ARROW unspecified\n - REPORT_TYPE_PROFILE_METADATA: REPORT_TYPE_PROFILE_METADATA contains metadata about the profile i.e. binaries, labels\n - REPORT_TYPE_EXPORT: REPORT_TYPE_EXPORT contains the profile as a downloadable file",
       "title": "ReportType is the type of report to return"
     },
     "metastorev1alpha1Function": {

diff --git a/parca.yaml b/parca.yaml
@@ -9,11 +9,11 @@ object_storage:
 # Scraping tends to produce a lot more data than Parca Agent CPU which
 # results in memory usage by the server.
 #
-# scrape_configs:
-#   - job_name: "parca"
-#     scrape_interval: "10s"
-#     static_configs:
-#       - targets: [ '127.0.0.1:7070' ]
+scrape_configs:
+  - job_name: "parca"
+    scrape_interval: "10s"
+    static_configs:
+      - targets: [ '127.0.0.1:7070' ]
 
 # Nested under the job config:
 #

diff --git a/pkg/parca/parca.go b/pkg/parca/parca.go
@@ -110,6 +110,8 @@ type Flags struct {
 
 	Storage FlagsStorage `embed:"" prefix:"storage-"`
 
+	ObjectStorageExport bool `default:"false" help:"Export profiles to object storage first before serving them via a signed URL."`
+
 	Symbolizer FlagsSymbolizer `embed:"" prefix:"symbolizer-"`
 
 	Debuginfo  FlagsDebuginfo  `embed:"" prefix:"debuginfo-"`
@@ -424,6 +426,9 @@ func Run(ctx context.Context, logger log.Logger, reg *prometheus.Registry, flags
 			debuginfoBucket,
 			debuginfodClients,
 		),
+		flags.ObjectStorageExport,
+		signedRequestsClient,
+		bucket,
 	)
 
 	t := telemetryservice.NewTelemetry(

diff --git a/pkg/query/columnquery.go b/pkg/query/columnquery.go
@@ -18,6 +18,9 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
+	"path"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -27,6 +30,7 @@ import (
 	"github.com/apache/arrow/go/v16/arrow/bitutil"
 	"github.com/apache/arrow/go/v16/arrow/math"
 	"github.com/apache/arrow/go/v16/arrow/memory"
+	"github.com/cespare/xxhash/v2"
 	"github.com/go-kit/log"
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/trace"
@@ -62,6 +66,14 @@ type SourceFinder interface {
 	SourceExists(ctx context.Context, ref *pb.SourceReference) (bool, error)
 }
 
+type SignedGetClient interface {
+	SignedGET(ctx context.Context, objectKey string, expiry time.Time) (string, error)
+}
+
+type ObjectStoreUploadClient interface {
+	Upload(ctx context.Context, objectKey string, r io.Reader) error
+}
+
 // ColumnQueryAPI is the read api interface for parca
 // It implements the proto/query/query.proto APIServer interface.
 type ColumnQueryAPI struct {
@@ -77,6 +89,13 @@ type ColumnQueryAPI struct {
 	converter          *parcacol.ArrowToProfileConverter
 
 	sourceFinder SourceFinder
+
+	// Whether to first upload an export to object storage and returning a
+	// signed URL to the client instead of serving the export directly within
+	// the API response.
+	objectStorgeExport bool
+	signedGetClient    SignedGetClient
+	objectStore        ObjectStoreUploadClient
 }
 
 func NewColumnQueryAPI(
@@ -87,6 +106,9 @@ func NewColumnQueryAPI(
 	mem memory.Allocator,
 	converter *parcacol.ArrowToProfileConverter,
 	sourceFinder SourceFinder,
+	objectStorageExport bool,
+	signedGetClient SignedGetClient,
+	objectStore ObjectStoreUploadClient,
 ) *ColumnQueryAPI {
 	return &ColumnQueryAPI{
 		logger:             logger,
@@ -97,6 +119,9 @@ func NewColumnQueryAPI(
 		mem:                mem,
 		converter:          converter,
 		sourceFinder:       sourceFinder,
+		objectStorgeExport: objectStorageExport,
+		signedGetClient:    signedGetClient,
+		objectStore:        objectStore,
 	}
 }
 
@@ -320,16 +345,24 @@ func (q *ColumnQueryAPI) Query(ctx context.Context, req *pb.QueryRequest) (*pb.Q
 		return nil, fmt.Errorf("filtering profile: %w", err)
 	}
 
+	data, err := req.MarshalVT()
+	if err != nil {
+		return nil, fmt.Errorf("marshaling request: %w", err)
+	}
+	requestKey := xxhash.Sum64(data)
+
 	return q.renderReport(
 		ctx,
 		p,
+		strconv.FormatUint(requestKey, 16),
 		req.GetReportType(),
 		req.GetNodeTrimThreshold(),
 		filtered,
 		groupBy,
 		req.GetSourceReference(),
 		source,
 		isDiff,
+		req.GetExportFormat(),
 	)
 }
 
@@ -474,17 +507,20 @@ func filterRecord(
 func (q *ColumnQueryAPI) renderReport(
 	ctx context.Context,
 	p profile.Profile,
+	requestKey string,
 	typ pb.QueryRequest_ReportType,
 	nodeTrimThreshold float32,
 	filtered int64,
 	groupBy []string,
 	sourceReference *pb.SourceReference,
 	source string,
 	isDiff bool,
+	exportFormat pb.Format,
 ) (*pb.QueryResponse, error) {
 	return RenderReport(
 		ctx,
 		q.tracer,
+		requestKey,
 		p,
 		typ,
 		nodeTrimThreshold,
@@ -496,12 +532,17 @@ func (q *ColumnQueryAPI) renderReport(
 		sourceReference,
 		source,
 		isDiff,
+		exportFormat,
+		q.objectStorgeExport,
+		q.signedGetClient,
+		q.objectStore,
 	)
 }
 
 func RenderReport(
 	ctx context.Context,
 	tracer trace.Tracer,
+	requestKey string,
 	p profile.Profile,
 	typ pb.QueryRequest_ReportType,
 	nodeTrimThreshold float32,
@@ -513,6 +554,10 @@ func RenderReport(
 	sourceReference *pb.SourceReference,
 	source string,
 	isDiff bool,
+	exportFormat pb.Format,
+	objectStorageExport bool,
+	signedGetClient SignedGetClient,
+	objectStore ObjectStoreUploadClient,
 ) (*pb.QueryResponse, error) {
 	ctx, span := tracer.Start(ctx, "renderReport")
 	span.SetAttributes(attribute.String("reportType", typ.String()))
@@ -609,6 +654,53 @@ func RenderReport(
 			Filtered: filtered,
 			Report:   &pb.QueryResponse_Pprof{Pprof: buf},
 		}, nil
+	case pb.QueryRequest_REPORT_TYPE_EXPORT:
+		var buf []byte
+
+		switch exportFormat {
+		case pb.Format_FORMAT_PPROF:
+			pp, err := GenerateFlatPprof(ctx, isDiff, p)
+			if err != nil {
+				return nil, status.Errorf(codes.Internal, "failed to generate pprof: %v", err.Error())
+			}
+
+			buf, err = SerializePprof(pp)
+			if err != nil {
+				return nil, status.Errorf(codes.Internal, "failed to generate pprof: %v", err.Error())
+			}
+		default:
+			return nil, status.Errorf(codes.InvalidArgument, "requested export format does not exist")
+		}
+
+		if objectStorageExport {
+			key := path.Join(requestKey, "pprof")
+			if err := objectStore.Upload(ctx, key, bytes.NewReader(buf)); err != nil {
+				return nil, status.Errorf(codes.Internal, "failed to upload export to object storage: %v", err.Error())
+			}
+
+			signedURL, err := signedGetClient.SignedGET(ctx, key, time.Now().Add(time.Hour))
+			if err != nil {
+				return nil, status.Errorf(codes.Internal, "failed to generate signed URL: %v", err.Error())
+			}
+
+			return &pb.QueryResponse{
+				Report: &pb.QueryResponse_Export{
+					Export: &pb.Export{
+						Format:  exportFormat,
+						Content: &pb.Export_Url{Url: &pb.URLExport{Url: signedURL}},
+					},
+				},
+			}, nil
+		}
+
+		return &pb.QueryResponse{
+			Report: &pb.QueryResponse_Export{
+				Export: &pb.Export{
+					Format:  exportFormat,
+					Content: &pb.Export_Inline{Inline: &pb.InlineExport{Content: buf}},
+				},
+			},
+		}, nil
 	case pb.QueryRequest_REPORT_TYPE_TOP:
 		op, err := converter.Convert(ctx, p)
 		if err != nil {