X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8ead06918642aa813655193d802ae82b9c363535..09cbdc3074b3f1e69c9c537875146f6da0a6ed8f:/lib/controller/federation/list.go

diff --git a/lib/controller/federation/list.go b/lib/controller/federation/list.go
index 414870d24f..329066d1dc 100644
--- a/lib/controller/federation/list.go
+++ b/lib/controller/federation/list.go
@@ -10,9 +10,10 @@ import (
 	"net/http"
 	"sort"
 	"sync"
+	"sync/atomic"
 
-	"git.curoverse.com/arvados.git/sdk/go/arvados"
-	"git.curoverse.com/arvados.git/sdk/go/httpserver"
+	"git.arvados.org/arvados.git/sdk/go/arvados"
+	"git.arvados.org/arvados.git/sdk/go/httpserver"
 )
 
 //go:generate go run generate.go
@@ -20,10 +21,13 @@ import (
 // CollectionList is used as a template to auto-generate List()
 // methods for other types; see generate.go.
 
-func (conn *Conn) CollectionList(ctx context.Context, options arvados.ListOptions) (arvados.CollectionList, error) {
+func (conn *Conn) generated_CollectionList(ctx context.Context, options arvados.ListOptions) (arvados.CollectionList, error) {
 	var mtx sync.Mutex
 	var merged arvados.CollectionList
+	var needSort atomic.Value
+	needSort.Store(false)
 	err := conn.splitListRequest(ctx, options, func(ctx context.Context, _ string, backend arvados.API, options arvados.ListOptions) ([]string, error) {
+		options.ForwardedFor = conn.cluster.ClusterID + "-" + options.ForwardedFor
 		cl, err := backend.CollectionList(ctx, options)
 		if err != nil {
 			return nil, err
@@ -32,8 +36,9 @@ func (conn *Conn) CollectionList(ctx context.Context, options arvados.ListOption
 		defer mtx.Unlock()
 		if len(merged.Items) == 0 {
 			merged = cl
-		} else {
+		} else if len(cl.Items) > 0 {
 			merged.Items = append(merged.Items, cl.Items...)
+			needSort.Store(true)
 		}
 		uuids := make([]string, 0, len(cl.Items))
 		for _, item := range cl.Items {
@@ -41,20 +46,32 @@ func (conn *Conn) CollectionList(ctx context.Context, options arvados.ListOption
 		}
 		return uuids, nil
 	})
-	sort.Slice(merged.Items, func(i, j int) bool { return merged.Items[i].UUID < merged.Items[j].UUID })
+	if needSort.Load().(bool) {
+		// Apply the default/implied order, "modified_at desc"
+		sort.Slice(merged.Items, func(i, j int) bool {
+			mi, mj := merged.Items[i].ModifiedAt, merged.Items[j].ModifiedAt
+			return mj.Before(mi)
+		})
+	}
+	if merged.Items == nil {
+		// Return empty results as [], not null
+		// (https://github.com/golang/go/issues/27589 might be
+		// a better solution in the future)
+		merged.Items = []arvados.Collection{}
+	}
 	return merged, err
 }
 
 // Call fn on one or more local/remote backends if opts indicates a
 // federation-wide list query, i.e.:
 //
-// * There is at least one filter of the form
-//   ["uuid","in",[a,b,c,...]] or ["uuid","=",a]
+//   - There is at least one filter of the form
+//     ["uuid","in",[a,b,c,...]] or ["uuid","=",a]
 //
-// * One or more of the supplied UUIDs (a,b,c,...) has a non-local
-//   prefix.
+//   - One or more of the supplied UUIDs (a,b,c,...) has a non-local
+//     prefix.
 //
-// * There are no other filters
+//   - There are no other filters
 //
 // (If opts doesn't indicate a federation-wide list query, fn is just
 // called once with the local backend.)
@@ -62,34 +79,46 @@ func (conn *Conn) CollectionList(ctx context.Context, options arvados.ListOption
 // fn is called more than once only if the query meets the following
 // restrictions:
 //
-// * Count=="none"
+//   - Count=="none"
 //
-// * Limit<0
+//   - Limit<0
 //
-// * len(Order)==0
+//   - len(Order)==0
 //
-// * Each filter must be either "uuid = ..." or "uuid in [...]".
+//   - Each filter is either "uuid = ..." or "uuid in [...]".
 //
-// * The maximum possible response size (total number of objects that
-//   could potentially be matched by all of the specified filters)
-//   exceeds the local cluster's response page size limit.
+//   - The maximum possible response size (total number of objects
+//     that could potentially be matched by all of the specified
+//     filters) exceeds the local cluster's response page size limit.
 //
 // If the query involves multiple backends but doesn't meet these
 // restrictions, an error is returned without calling fn.
 //
 // Thus, the caller can assume that either:
 //
-// * splitListRequest() returns an error, or
+//   - splitListRequest() returns an error, or
 //
-// * fn is called exactly once, or
+//   - fn is called exactly once, or
 //
-// * fn is called more than once, with options that satisfy the above
-//   restrictions.
+//   - fn is called more than once, with options that satisfy the above
+//     restrictions.
 //
 // Each call to fn indicates a single (local or remote) backend and a
 // corresponding options argument suitable for sending to that
 // backend.
 func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions, fn func(context.Context, string, arvados.API, arvados.ListOptions) ([]string, error)) error {
+
+	if opts.BypassFederation || opts.ForwardedFor != "" {
+		// Client requested no federation.  Pass through.
+		_, err := fn(ctx, conn.cluster.ClusterID, conn.local, opts)
+		return err
+	}
+	if opts.ClusterID != "" {
+		// Client explicitly selected cluster
+		_, err := fn(ctx, conn.cluster.ClusterID, conn.chooseBackend(opts.ClusterID), opts)
+		return err
+	}
+
 	cannotSplit := false
 	var matchAllFilters map[string]bool
 	for _, f := range opts.Filters {
@@ -139,6 +168,13 @@ func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions
 		}
 	}
 
+	if matchAllFilters == nil {
+		// Not filtering by UUID at all; just query the local
+		// cluster.
+		_, err := fn(ctx, conn.cluster.ClusterID, conn.local, opts)
+		return err
+	}
+
 	// Collate UUIDs in matchAllFilters by remote cluster ID --
 	// e.g., todoByRemote["aaaaa"]["aaaaa-4zz18-000000000000000"]
 	// will be true -- and count the total number of UUIDs we're
@@ -158,28 +194,27 @@ func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions
 		}
 	}
 
-	if len(todoByRemote) > 1 {
-		if cannotSplit {
-			return httpErrorf(http.StatusBadRequest, "cannot execute federated list query: each filter must be either 'uuid = ...' or 'uuid in [...]'")
-		}
-		if opts.Count != "none" {
-			return httpErrorf(http.StatusBadRequest, "cannot execute federated list query unless count==\"none\"")
-		}
-		if opts.Limit >= 0 || opts.Offset != 0 || len(opts.Order) > 0 {
-			return httpErrorf(http.StatusBadRequest, "cannot execute federated list query with limit, offset, or order parameter")
-		}
-		if max := conn.cluster.API.MaxItemsPerResponse; nUUIDs > max {
-			return httpErrorf(http.StatusBadRequest, "cannot execute federated list query because number of UUIDs (%d) exceeds page size limit %d", nUUIDs, max)
-		}
-		selectingUUID := false
-		for _, attr := range opts.Select {
-			if attr == "uuid" {
-				selectingUUID = true
-			}
-		}
-		if opts.Select != nil && !selectingUUID {
-			return httpErrorf(http.StatusBadRequest, "cannot execute federated list query with a select parameter that does not include uuid")
-		}
+	if len(todoByRemote) == 0 {
+		return nil
+	}
+	if len(todoByRemote) == 1 && todoByRemote[conn.cluster.ClusterID] != nil {
+		// All UUIDs are local, so proxy a single request. The
+		// generic case has some limitations (see below) which
+		// we don't want to impose on local requests.
+		_, err := fn(ctx, conn.cluster.ClusterID, conn.local, opts)
+		return err
+	}
+	if cannotSplit {
+		return httpErrorf(http.StatusBadRequest, "cannot execute federated list query: each filter must be either 'uuid = ...' or 'uuid in [...]'")
+	}
+	if opts.Count != "none" {
+		return httpErrorf(http.StatusBadRequest, "cannot execute federated list query unless count==\"none\"")
+	}
+	if (opts.Limit >= 0 && opts.Limit < int64(nUUIDs)) || opts.Offset != 0 || len(opts.Order) > 0 {
+		return httpErrorf(http.StatusBadRequest, "cannot execute federated list query with limit (%d) < nUUIDs (%d), offset (%d) > 0, or order (%v) parameter", opts.Limit, nUUIDs, opts.Offset, opts.Order)
+	}
+	if max := conn.cluster.API.MaxItemsPerResponse; nUUIDs > max {
+		return httpErrorf(http.StatusBadRequest, "cannot execute federated list query because number of UUIDs (%d) exceeds page size limit %d", nUUIDs, max)
 	}
 
 	ctx, cancel := context.WithCancel(ctx)
@@ -202,6 +237,12 @@ func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions
 				return
 			}
 			remoteOpts := opts
+			if remoteOpts.Select != nil {
+				// We always need to select UUIDs to
+				// use the response, even if our
+				// caller doesn't.
+				remoteOpts.Select = append([]string{"uuid"}, remoteOpts.Select...)
+			}
 			for len(todo) > 0 {
 				if len(batch) > len(todo) {
 					// Reduce batch to just the todo's
@@ -214,7 +255,7 @@ func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions
 
 				done, err := fn(ctx, clusterID, backend, remoteOpts)
 				if err != nil {
-					errs <- err
+					errs <- httpErrorf(http.StatusBadGateway, "%s", err.Error())
 					return
 				}
 				progress := false
@@ -224,8 +265,13 @@ func (conn *Conn) splitListRequest(ctx context.Context, opts arvados.ListOptions
 						delete(todo, uuid)
 					}
 				}
-				if !progress {
-					errs <- httpErrorf(http.StatusBadGateway, "cannot make progress in federated list query: cluster %q returned none of the requested UUIDs", clusterID)
+				if len(done) == 0 {
+					// Zero items == no more
+					// results exist, no need to
+					// get another page.
+					break
+				} else if !progress {
+					errs <- httpErrorf(http.StatusBadGateway, "cannot make progress in federated list query: cluster %q returned %d items but none had the requested UUIDs", clusterID, len(done))
 					return
 				}
 			}