func (spec specLocalLogic) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "logic-link", } // search for existing link re := g.OutWith(src.ID, q.Qbe(system.EType("logic-link"), "path", spec.Path)) if len(re) == 1 { // TODO don't set the path prop again, it's the unique id...meh, same question here w/uniqueness as above success = true e = re[0] return } // no existing link found, search for proc directly envid, _, _ := findEnv(g, src) rv := g.PredecessorsWith(envid, q.Qbv(system.VType("logic-state"), "path", spec.Path)) if len(rv) == 1 { success = true e.Target = rv[0].ID } return }
func (spec specParentDataset) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "dataset-gateway", } e.Props = e.Props.Set("name", system.Property{MsgSrc: mid, Value: spec.Name}) // check for existing link - there can be only be one re := g.OutWith(src.ID, q.Qbe(system.EType("dataset-gateway"))) if len(re) == 1 { success = true e = re[0] // TODO semantics should preclude this from being able to change, but doing it dirty means force-setting it anyway for now } else { // no existing link found; search for proc directly envid, _, _ := findEnv(g, src) rv := g.PredecessorsWith(envid, q.Qbv(system.VType("parent-dataset"), "name", spec.Name)) if len(rv) != 0 { // >1 shouldn't be possible success = true e.Target = rv[0].ID } } return }
func (spec EnvLink) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { _, e, success = findEnv(g, src) // Whether we find a match or not, have to merge in the EnvLink e.Props = maputil.FillPropMap(mid, false, pp("hostname", spec.Address.Hostname), pp("ipv4", spec.Address.Ipv4), pp("ipv6", spec.Address.Ipv6), pp("nick", spec.Nick), ) // If we already found the matching edge, bail out now if success { return } rv := g.VerticesWith(q.Qbv(system.VType("environment"))) for _, vt := range rv { // TODO this'll be cross-package eventually - reorg needed if maputil.AnyMatch(e.Props, vt.Vertex.Properties, "nick", "hostname", "ipv4", "ipv6") { success = true e.Target = vt.ID break } } return }
func (spec specGitCommitParent) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "parent-commit", } re := g.OutWith(src.ID, q.Qbe(system.EType("parent-commit"), "pnum", spec.ParentNum)) if len(re) > 0 { success = true e.Target = re[0].Target e.Props = re[0].Props // FIXME evidence of a problem here - since we're using pnum as the deduping identifier, there's no // way it could also sensibly change its MsgSrc value. This is very much a product of the intensional/extensional // identity problem: what does it mean to have the identifying data change? is it now a new thing? was it the old thing, // and it underwent a transition into the new thing? or is there no distinction between the old and new thing? e.Props = e.Props.Set("sha1", system.Property{MsgSrc: mid, Value: spec.Sha1}) e.ID = re[0].ID } else { rv := g.VerticesWith(q.Qbv(system.VType("commit"), "sha1", spec.Sha1)) if len(rv) == 1 { success = true e.Target = rv[0].ID e.Props = e.Props.Set("pnum", system.Property{MsgSrc: mid, Value: spec.ParentNum}) e.Props = e.Props.Set("sha1", system.Property{MsgSrc: mid, Value: spec.Sha1}) } } return }
func (spec specUnixDomainListener) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { // check for existing edge; this one is quite straightforward re := g.OutWith(src.ID, q.Qbe(system.EType("listening"), "type", "unix", "path", spec.Path)) if len(re) == 1 { return re[0], true } e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "listening", } e.Props = e.Props.Set("path", system.Property{MsgSrc: mid, Value: spec.Path}) envid, _, hasenv := findEnv(g, src) if hasenv { rv := g.PredecessorsWith(envid, q.Qbv(system.VType("comm"), "type", "unix", "path", spec.Path)) if len(rv) == 1 { success = true e.Target = rv[0].ID } } return }
func BenchmarkMergeMessageOneAndTwo(b *testing.B) { var g system.CoreGraph = &coreGraph{vtuples: ps.NewMap()} for i := 0; i < b.N; i++ { g.Merge(0, msgs[0].UnificationForm()) g.Merge(0, msgs[1].UnificationForm()) } }
func commitUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { candidates := g.VerticesWith(q.Qbv(system.VType("commit"), "sha1", u.Vertex().Properties()["sha1"])) if len(candidates) > 0 { // there can be only one return candidates[0].ID } return 0 }
func parentDatasetUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { edge, success := u.ScopingSpecs()[0].(EnvLink).Resolve(g, 0, emptyVT(u.Vertex())) if !success { // FIXME scoping edge resolution failure does not mean no match - there could be an orphan return 0 } props := u.Vertex().Properties() return findMatchingEnvId(g, edge, g.VerticesWith(q.Qbv(system.VType("parent-dataset"), "path", props["path"], "name", props["name"]))) }
func processUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { // only one scoping edge - the envlink edge, success := u.ScopingSpecs()[0].(EnvLink).Resolve(g, 0, emptyVT(u.Vertex())) if !success { // FIXME scoping edge resolution failure does not mean no match - there could be an orphan return 0 } return findMatchingEnvId(g, edge, g.VerticesWith(q.Qbv(system.VType("process"), "pid", u.Vertex().Properties()["pid"]))) }
func findEnvironment(g system.CoreGraph, props ps.Map) (envid uint64, success bool) { rv := g.VerticesWith(q.Qbv(system.VType("environment"))) for _, vt := range rv { if maputil.AnyMatch(props, vt.Vertex.Props(), "hostname", "ipv4", "ipv6", "nick") { return vt.ID, true } } return }
func envUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { matches := g.VerticesWith(q.Qbv(system.VType("environment"))) for _, e := range matches { if maputil.AnyMatch(e.Vertex.Properties, u.Vertex().Properties(), "hostname", "ipv4", "ipv6") { return e.ID } } return 0 }
func findMatchingEnvId(g system.CoreGraph, edge system.StdEdge, vtv system.VertexTupleVector) uint64 { for _, candidate := range vtv { for _, edge2 := range g.OutWith(candidate.ID, q.Qbe(system.EType("envlink"))) { if edge2.Target == edge.Target { return candidate.ID } } } return 0 }
func pkgYumUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { props := u.Vertex().Properties() vtv := g.VerticesWith(q.Qbv(system.VType("pkg-yum"), "name", props["name"], "version", props["version"], "arch", props["arch"], "epoch", props["epoch"], )) if len(vtv) > 0 { return vtv[0].ID } return 0 }
func graphToJSON(g system.CoreGraph) ([]byte, error) { var vertices []interface{} for _, v := range g.VerticesWith(q.Qbv(system.VTypeNone)) { vertices = append(vertices, v.Flat()) } // TODO use something that lets us write to a reusable byte buffer instead return json.Marshal(struct { Id uint64 `json:"id"` Vertices []interface{} `json:"vertices"` }{ Id: g.MsgID(), Vertices: vertices, }) }
// Searches the given vertex's out-edges to find its environment's vertex id. // // Also conveniently initializes a StandardEdge to the standard zero-state for an envlink. func findEnv(g system.CoreGraph, vt system.VertexTuple) (vid uint64, edge system.StdEdge, success bool) { edge = system.StdEdge{ Source: vt.ID, Props: ps.NewMap(), EType: "envlink", } if vt.ID != 0 { re := g.OutWith(vt.ID, q.Qbe(system.EType("envlink"))) if len(re) == 1 { vid, edge, success = re[0].Target, re[0], true } } return }
func (spec DataAlpha) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { // TODO this makes a loop...are we cool with that? success = true // impossible to fail here e = system.StdEdge{ Source: src.ID, Target: src.ID, Props: ps.NewMap(), EType: "data-provenance", } re := g.OutWith(src.ID, q.Qbe(system.EType("data-provenance"))) if len(re) == 1 { e = re[0] } return }
func datasetUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { vtv := g.VerticesWith(q.Qbv(system.VType("dataset"), "name", u.Vertex().Properties()["name"])) if len(vtv) == 0 { return 0 } spec := u.ScopingSpecs()[0].(specDatasetHierarchy) el, success := spec.Environment.Resolve(g, 0, emptyVT(u.Vertex())) // FIXME scoping edge resolution failure does not mean no match - there could be an orphan if success { for _, vt := range vtv { if id := findMatchingEnvId(g, el, g.SuccessorsWith(vt.ID, q.Qbe(system.EType("dataset-hierarchy")))); id != 0 { return vt.ID } } } return 0 }
func findDataset(g system.CoreGraph, envid uint64, name []string) (id uint64, success bool) { // first time through use the parent type vtype := system.VType("parent-dataset") id = envid var n string for len(name) > 0 { n, name = name[0], name[1:] rv := g.PredecessorsWith(id, q.Qbv(vtype, "name", n)) vtype = "dataset" if len(rv) != 1 { return 0, false } id = rv[0].ID } return id, true }
func (spec DataProvenance) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { // FIXME this presents another weird case where "success" is not binary. We *could* // find an already-existing data-provenance edge, but then have some net-addr params // change which cause it to fail to resolve to an environment. If we call that successful, // then it won't try to resolve again later...though, hm, just call it unsuccessful and // then try again one more time. Maybe it is fine. THINK IT THROUGH. e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "data-provenance", } e.Props = assignAddress(mid, spec.Address, e.Props, false) re := g.OutWith(src.ID, q.Qbe(system.EType("data-provenance"))) if len(re) == 1 { reresolve := maputil.AnyMatch(e.Props, re[0].Props, "hostname", "ipv4", "ipv6") e = re[0] if spec.SnapTime != "" { e.Props = e.Props.Set("snap-time", system.Property{MsgSrc: mid, Value: spec.SnapTime}) } if reresolve { e.Props = assignAddress(mid, spec.Address, e.Props, true) } else { return e, true } } envid, found := findEnvironment(g, e.Props) if !found { // TODO returning this already-modified edge necessitates that the core system // disregard 'failed' edges. which should be fine, that should be a guarantee return e, false } e.Target, success = findDataset(g, envid, spec.Dataset) return }
func (spec specCommit) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "version", } e.Props = e.Props.Set("sha1", system.Property{MsgSrc: mid, Value: spec.Sha1}) re := g.OutWith(src.ID, q.Qbe(system.EType("version"))) if len(re) > 0 { sha1, _ := re[0].Props.Lookup("sha1") e.ID = re[0].ID // FIXME setting the id to non-0 AND failing is currently unhandled if sha1.(system.Property).Value == spec.Sha1 { success = true e.Target = re[0].Target } else { rv := g.VerticesWith(q.Qbv(system.VType("commit"), "sha1", spec.Sha1)) if len(rv) == 1 { success = true e.Target = rv[0].ID } } } else { rv := g.VerticesWith(q.Qbv(system.VType("commit"), "sha1", spec.Sha1)) if len(rv) == 1 { success = true e.Target = rv[0].ID } } return }
func commUnify(g system.CoreGraph, u system.UnifyInstructionForm) uint64 { // only one scoping edge - the envlink edge, success := u.ScopingSpecs()[0].(EnvLink).Resolve(g, 0, emptyVT(u.Vertex())) if !success { // FIXME scoping edge resolution failure does not mean no match - there could be an orphan return 0 } vp := u.Vertex().Properties() typ, _ := vp["type"] path, haspath := vp["path"] if haspath { return findMatchingEnvId(g, edge, g.VerticesWith(q.Qbv(system.VType("comm"), "type", typ, "path", path))) } else { port, _ := vp["port"] return findMatchingEnvId(g, edge, g.VerticesWith(q.Qbv(system.VType("comm"), "type", typ, "port", port))) } }
func init() { startTime := time.Now().UTC() expvar.Publish("Uptime", expvar.Func(func() interface{} { return int64(time.Since(startTime)) })) expvar.Publish("Goroutines", expvar.Func(func() interface{} { return runtime.NumGoroutine() })) // subscribe to the broker in order to report data about current graph c := broker.Get().Subscribe() // Instantiate a real, empty graph to ensure the interface type is never nil when it might be called var g system.CoreGraph = represent.NewGraph() go func() { for latest := range c { g = latest } }() expvar.Publish("MsgId", expvar.Func(func() interface{} { return g.MsgId() })) go func() { log.Println(http.ListenAndServe("localhost:6060", nil)) }() }
// Generates a .dot-format representation of the given CoreGraph, suitable for // rendering into output by graphviz (or other utilities). func GenerateDot(g system.CoreGraph) []byte { buf := new(bytes.Buffer) // begin the graph buf.WriteString("digraph{\n") buf.WriteString("fontsize=16") // first, write all vertices for _, v := range g.VerticesWith(q.Qbv()) { lbltype := "label" var props string switch v.Vertex.Typ() { case "environment": props = "\tshape=house,style=filled,fillcolor=orange,fontsize=20\n" case "logic-state": props = "\tshape=box3d,style=filled,fillcolor=purple,fontcolor=white,fontsize=18,\n" case "process": props = "\tshape=oval,style=filled,fillcolor=green,\n" case "dataset", "parent-dataset": props = "\tshape=folder,style=filled,fillcolor=brown,fontcolor=white,\n" case "comm": props = "\tshape=doubleoctagon,style=filled,fillcolor=cyan,\n" case "git-commit": props = "\tshape=box,style=filled,fillcolor=grey\n" case "git-tag", "git-branch": props = "\tshape=cds,margin=\"0.22,0.22\",\n" case "test-result": props = "\tshape=note\n" } buf.WriteString(fmt.Sprintf( "\t\"v%d\" [%s%s=\"id: %d\nvtype: %s", v.ID, props, lbltype, v.ID, v.Vertex.Typ())) v.Vertex.Props().ForEach(func(k string, val ps.Any) { prop := val.(system.Property) var format string switch pv := prop.Value.(type) { case []byte, [20]byte: format = "%x" case int, int64, int32, int16, int8, uint, uint64, uint32, uint16, uint8: format = "%d" case string: prop.Value = strings.Trim(strconv.QuoteToASCII(pv), `"`) format = "%s" default: format = "%s" } buf.WriteString(fmt.Sprintf( "\n%s: "+format+" (%d)", k, prop.Value, prop.MsgSrc)) }) buf.WriteString("\"\n") buf.WriteString("];\n") } // pass through a second time to write all edges for _, v := range g.VerticesWith(q.Qbv()) { v.OutEdges.ForEach(func(k string, val ps.Any) { edge := val.(system.StdEdge) buf.WriteString(fmt.Sprintf( "\t\"v%d\" -> \"v%d\" [\n\tlabel=\"id: %d\netype: %s", edge.Source, edge.Target, edge.ID, edge.EType)) edge.Props.ForEach(func(k2 string, val2 ps.Any) { prop := val2.(system.Property) var format string switch pv := prop.Value.(type) { case []byte: format = "%x" case int, int64, int32, int16, int8, uint, uint64, uint32, uint16, uint8: format = "%d" case string: prop.Value = strings.Trim(strconv.QuoteToASCII(pv), `"`) format = "%s" default: format = "%s" } buf.WriteString(fmt.Sprintf( "\n%s: "+format+" (%d)", k2, prop.Value, prop.MsgSrc)) }) buf.WriteString("\"\n") switch edge.EType { case "envlink": buf.WriteString("\tstyle=dashed\n") } buf.WriteString("];\n") }) } // close out the graph buf.WriteString("}\n") return buf.Bytes() }
func (spec DataLink) Resolve(g system.CoreGraph, mid uint64, src system.VertexTuple) (e system.StdEdge, success bool) { e = system.StdEdge{ Source: src.ID, Props: ps.NewMap(), EType: "datalink", } // DataLinks have a 'name' field that is expected to be unique for the source, if present if spec.Name != "" { // TODO 'name' is a traditional unique key; a change in it inherently denotes a new edge. how to handle this? // FIXME this approach just always updates the mid, which is weird? e.Props = e.Props.Set("name", system.Property{MsgSrc: mid, Value: spec.Name}) re := g.OutWith(src.ID, q.Qbe(system.EType("datalink"), "name", spec.Name)) if len(re) == 1 { success = true e = re[0] } } if spec.Type != "" { e.Props = e.Props.Set("type", system.Property{MsgSrc: mid, Value: spec.Type}) } if spec.Subset != "" { e.Props = e.Props.Set("subset", system.Property{MsgSrc: mid, Value: spec.Subset}) } if spec.Interaction != "" { e.Props = e.Props.Set("interaction", system.Property{MsgSrc: mid, Value: spec.Interaction}) } // Special bits: if we have ConnUnix data, eliminate ConnNet data, and vice-versa. var isLocal bool if spec.ConnUnix.Path != "" { isLocal = true e.Props = e.Props.Set("path", system.Property{MsgSrc: mid, Value: spec.ConnUnix.Path}) e.Props = e.Props.Delete("hostname") e.Props = e.Props.Delete("ipv4") e.Props = e.Props.Delete("ipv6") e.Props = e.Props.Delete("port") e.Props = e.Props.Delete("proto") } else { e.Props = e.Props.Set("port", system.Property{MsgSrc: mid, Value: spec.ConnNet.Port}) e.Props = e.Props.Set("proto", system.Property{MsgSrc: mid, Value: spec.ConnNet.Proto}) // can only be one of hostname, ipv4 or ipv6 if spec.ConnNet.Hostname != "" { e.Props = e.Props.Set("hostname", system.Property{MsgSrc: mid, Value: spec.ConnNet.Hostname}) } else if spec.ConnNet.Ipv4 != "" { e.Props = e.Props.Set("ipv4", system.Property{MsgSrc: mid, Value: spec.ConnNet.Ipv4}) } else { e.Props = e.Props.Set("ipv6", system.Property{MsgSrc: mid, Value: spec.ConnNet.Ipv6}) } } if success { return } var sock system.VertexTuple var rv system.VertexTupleVector // just for reuse // If net, must scan; if local, a bit easier. if !isLocal { // First, find the environment vertex rv = g.VerticesWith(q.Qbv(system.VType("environment"))) var envid uint64 for _, vt := range rv { if maputil.AnyMatch(e.Props, vt.Vertex.Properties, "hostname", "ipv4", "ipv6") { envid = vt.ID break } } // No matching env found, bail out if envid == 0 { return } // Now, walk the environment's edges to find the vertex representing the port rv = g.PredecessorsWith(envid, q.Qbv(system.VType("comm"), "type", "port", "port", spec.ConnNet.Port).And(q.Qbe(system.EType("envlink")))) if len(rv) != 1 { return } sock = rv[0] // With sock in hand, now find its proc rv = g.PredecessorsWith(sock.ID, q.Qbe(system.EType("listening"), "proto", spec.ConnNet.Proto).And(q.Qbv(system.VType("process")))) if len(rv) != 1 { // TODO could/will we ever allow >1? return } } else { envid, _, exists := findEnv(g, src) if !exists { // this is would be a pretty weird case return } // Walk the graph to find the vertex representing the unix socket rv = g.PredecessorsWith(envid, q.Qbv(system.VType("comm"), "path", spec.ConnUnix.Path).And(q.Qbe(system.EType("envlink")))) if len(rv) != 1 { return } sock = rv[0] // With sock in hand, now find its proc rv = g.PredecessorsWith(sock.ID, q.Qbv(system.VType("process")).And(q.Qbe(system.EType("listening")))) if len(rv) != 1 { // TODO could/will we ever allow >1? return } } rv = g.SuccessorsWith(rv[0].ID, q.Qbv(system.VType("parent-dataset"))) // FIXME this absolutely could be more than 1 if len(rv) != 1 { return } dataset := rv[0] // if the spec indicates a subset, find it if spec.Subset != "" { rv = g.PredecessorsWith(rv[0].ID, q.Qbv(system.VType("dataset"), "name", spec.Subset).And(q.Qbe(system.EType("dataset-hierarchy")))) if len(rv) != 1 { return } dataset = rv[0] } // FIXME only recording the final target id is totally broken; see https://github.com/pipeviz/pipeviz/issues/37 // Aaaand we found our target. success = true e.Target = dataset.ID return }