featurebase/querycontext/visualize.go
Seebs 0c94a53a73 initial implementation of QueryContext design
This is living in a subdirectory for now so we can have better
turnaround time on tests and not have to build everything else
along with it.

This covers the logic that we can have *without* actually using
databases or the filesystem in any way, just to provide a framework
that lets us validate the logic handling overlapping queries.

The overall purpose of this is to prevent deadlocks, by ensuring
that database locks are only taken when we have already proven
that they are available. In short, the QueryContext preregisters
its "scope" -- the set of things it may want to lock. The operation
of creating the QueryContext can block, but it blocks with no
database locks held. Once it is unblocked, the scope it has reported
is now considered unavailable, and no other QueryContext using any
overlapping scope can complete creation until this QueryContext
completes. While it's running, the QueryContext can't request write
access to anything outside its scope. Thus, once created, a
QueryContext can always proceed, without being blocked, until it's
done.

Note that this does not fully address multi-node behaviors;
once you have a QueryContext blocking things, you need to not
make queries to other nodes that could be blocked in turn by those
nodes. In short, no write queries to other nodes while holding a
write-type QueryContext on the local node, because if two nodes
do that to each other at once, they can both be blocked.

We believe RBF is currently designed such that read-only accesses
don't block progress on writes, so non-write access doesn't
create problems.

We also have some code to allow us to create dot-format output
from the components of this system, which is mostly intended to
be a debugging tool.

(cherry picked from commit e3d137f29c)
2022-11-15 11:22:17 -08:00

396 lines
9.5 KiB
Go

package querycontext
import (
"fmt"
"io"
)
// Rendering time!
// We want to be able to render an rbfTxStore, or a QueryContext, or the whole system.
//
// Here's the things we have:
// rbfDBQueryContexts
// rbfTxStore
// rbfTxWrappers
// rbfQueryContext
// rbfQueryRead
// rbfQueryWrite
// rbf.DB
// rbf.Tx
// QueryScope
//
// Relationships that exist:
// rbfTxStore -> map[dbKey]rbfDBQueryContexts -> {one rbf.DB, map[rbfTxWrappers]rbfQueryContext}
// rbfQueryContext -> many rbfTxWrappers -> {one rbf.Tx, one rbfDBQueryContext, map[txKey]QueryRead}
// rbfQueryRead -> rbfTxWrappers
//
// For a TxStore, it's useful to graph the set of rbfDBQueryContexts it
// has (which represent live databases), and the QueryContexts associated
// with them. For a QueryContext, it's useful to graph the set of
// rbfTxWrappers it has, and also its scope.
//
// It is also interesting to graph the whole thing. There should always
// be a round trip in these relations; an rbfTxWrapper always maps to
// an rbfDBQueryContext, which should always have a reference back to
// the rbfQueryContext that owns the rbfTxWrapper. Similarly, if
// you have an rbfDBQueryContext, every rbfTxWrapper it references
// has a Tx against the associated DB.
// colors: ['#d7191c','#fdae61','#ffffbf','#abd9e9','#2c7bb6'] is a
// colorblind-friendly palette obtained from ColorBrewer. We assign
// meanings:
const (
colorBlack = "#000000"
colorWhite = "#ffffff"
colorWrite = "#fdae61"
colorRead = "#abd9e9"
colorError = "#d7191c"
colorUnused1 = "#ffffbf"
colorUnused2 = "##2c7bb6"
)
var colorWritable = map[bool]string{
false: colorRead,
true: colorWrite,
}
type dotNode interface {
writeNode(io.Writer)
addTo(*dotGraph)
dotId() string
dotClass() string
}
type dotEdge interface {
writeEdge(io.Writer, *dotGraph)
}
// dotGraph stores a list of nodes and edges it's found so far.
// we also have a queue. the reason is that if you don't, descent
// is depth-first, and this can lead to finding a node with very
// little depth left before you would have found it with a higher
// depth, and potentially not recording its edges.
type dotGraph struct {
depth int
queue []dotNode
edges []dotEdge
nodes map[dotNode]string
orderedNodes []dotNode
clusters map[string][]dotNode
}
func (dg *dotGraph) enqueue(n dotNode) {
if _, ok := dg.nodes[n]; ok {
return
}
dg.queue = append(dg.queue, n)
}
// build starts with the enqueued roots and processes
// them, looking for new things, depth times.
func (dg *dotGraph) build(depth int) {
if dg.nodes == nil {
dg.nodes = make(map[dotNode]string)
}
dg.depth = depth
for len(dg.queue) > 0 && dg.depth > 0 {
working := dg.queue
dg.queue = []dotNode{}
for _, node := range working {
node.addTo(dg)
}
dg.depth--
}
}
// consider checks whether we're sufficiently done with this item, which
// means either we've already seen it, or we haven't but we're out of depth.
func (dg *dotGraph) consider(n dotNode) bool {
if _, ok := dg.nodes[n]; ok {
return true
}
dg.nodes[n] = n.dotId()
// maintain a list in the order we saw them, so items at the top of
// a given list will correspond to items at the top of the next list too
dg.orderedNodes = append(dg.orderedNodes, n)
return dg.depth < 1
}
var nodeQualities = map[string][2]map[string]string{
"rdqc": {{"rank": "same"}, {"shape": "cylinder"}},
"txw": {{"rank": "same"}, {"shape": "folder"}},
"qrw": {{"rank": "same"}, {"shape": "note"}},
"txstore": {{"rank": "same"}, {"shape": "hexagon"}},
"rqc": {{"rank": "same"}, {"shape": "box3d"}},
}
var nodeOrder = []string{"txstore", "rdqc", "txw", "qrw", "rqc"}
func (dg *dotGraph) WriteClass(w io.Writer, class string) error {
nodes := dg.clusters[class]
if len(nodes) == 0 {
return nil
}
fmt.Fprintf(w, "\tsubgraph %s {\n", class)
if qualities, ok := nodeQualities[class]; ok {
for k, v := range qualities[0] {
fmt.Fprintf(w, "\t\t%s=%q\n", k, v)
}
if len(qualities[1]) > 0 {
fmt.Fprintf(w, "\t\tnode [")
for k, v := range qualities[1] {
fmt.Fprintf(w, "%s=%q", k, v)
}
fmt.Fprintln(w, "]")
}
}
for _, node := range nodes {
fmt.Fprintf(w, "\t\t\"%s\" ", dg.nodes[node])
node.writeNode(w)
fmt.Fprintln(w, "")
}
fmt.Fprintf(w, "\t}\n")
return nil
}
func (dg *dotGraph) Write(w io.Writer) error {
w.Write([]byte(`
digraph g {
rankdir = "LR"
newrank = true
concentrate = true
`))
dg.clusters = make(map[string][]dotNode)
classes := make(map[string]struct{})
for _, node := range dg.orderedNodes {
class := node.dotClass()
dg.clusters[class] = append(dg.clusters[class], node)
classes[class] = struct{}{}
}
for _, class := range nodeOrder {
err := dg.WriteClass(w, class)
if err != nil {
return err
}
delete(classes, class)
}
if len(classes) > 0 {
return fmt.Errorf("unordered classes: %v", classes)
}
for _, edge := range dg.edges {
edge.writeEdge(w, dg)
}
fmt.Fprintln(w, "}")
return nil
}
// pointerEdge is a naive edge that just represents two pointer-shaped things
type pointerEdge struct {
from, to dotNode
label string
style string
}
func (d *pointerEdge) writeEdge(w io.Writer, dg *dotGraph) {
var style string
if d.style != "" {
style = fmt.Sprintf(" style=%q", d.style)
}
fmt.Fprintf(w, " %q -> %q [label=%q%s]\n", dg.nodes[d.from], dg.nodes[d.to], d.label, style)
}
func (dg *dotGraph) addEdge(from, to dotNode, extra ...string) {
var label, style string
switch len(extra) {
case 2:
style = extra[1]
fallthrough
case 1:
label = extra[0]
}
edge := &pointerEdge{from, to, label, style}
dg.edges = append(dg.edges, edge)
}
// RBF backend to Graphviz glue
func (r *rbfTxStore) addTo(dg *dotGraph) {
if dg.consider(r) {
return
}
for _, dbqc := range r.dbs {
dg.enqueue(dbqc)
dg.addEdge(r, dbqc)
}
}
func (r *rbfDBQueryContexts) addTo(dg *dotGraph) {
if dg.consider(r) {
return
}
for tx, qc := range r.queryContexts {
dg.enqueue(qc)
dg.enqueue(tx)
dg.addEdge(r, qc)
dg.addEdge(r, tx, qc.name)
}
}
func (r *rbfQueryContext) addTo(dg *dotGraph) {
if dg.consider(r) {
return
}
dg.addEdge(r, r.txStore, "", "dashed")
for _, wrappers := range r.queries {
dg.enqueue(wrappers)
dg.addEdge(r, wrappers)
}
}
func (r *rbfTxWrappers) addTo(dg *dotGraph) {
if dg.consider(r) {
return
}
dg.enqueue(r.db)
dg.addEdge(r, r.db, "", "dashed")
for _, reader := range r.queries {
switch v := reader.(type) {
case *rbfQueryRead:
dg.enqueue(v)
dg.addEdge(r, v)
case *rbfQueryWrite:
dg.enqueue(v)
dg.addEdge(r, v)
default:
unk := &unknownEntry{v}
dg.enqueue(unk)
dg.addEdge(r, unk)
}
}
}
func (r *rbfTxStore) dotClass() string {
return "txstore"
}
func (r *rbfTxStore) dotId() string {
return fmt.Sprintf("txstore-%p", r)
}
func (r *rbfTxStore) writeNode(w io.Writer) {
fmt.Fprintf(w, `[label=<rbfTxStore<BR/>[<FONT POINT-SIZE="12" FACE="courier">%s</FONT>]>]`, r.rootPath)
}
func (r *rbfDBQueryContexts) dotClass() string {
return "rdqc"
}
func (r *rbfDBQueryContexts) dotId() string {
return fmt.Sprintf("rdqc-%p", r)
}
func (r *rbfDBQueryContexts) writeNode(w io.Writer) {
var locked string
var maybeError = colorBlack
if r.lockCheck.TryLock() {
r.lockCheck.Unlock()
} else {
locked = fmt.Sprintf("<BR/><FONT COLOR=%q>[LOCKED]</FONT>", colorWrite)
}
// We'll want something like this when we get to adding the RBF backend.
if false {
if r.db == nil {
maybeError = colorError
}
}
fmt.Fprintf(w, `[label=<rbfDBQueryContexts<BR/><FONT POINT-SIZE="12" FACE="courier">%s</FONT><BR/>[DB %p]%s> color=%q]`, r.key, r.db, locked, maybeError)
}
func (r *rbfTxWrappers) dotClass() string {
return "txw"
}
func (r *rbfTxWrappers) dotId() string {
return fmt.Sprintf("txw-%p", r)
}
func (r *rbfTxWrappers) writeNode(w io.Writer) {
var write string
if r.writeTx {
write = "write "
}
fmt.Fprintf(w, `[label=<rbfTxWrappers<BR/><FONT POINT-SIZE="12" FACE="courier">%s</FONT><BR/>[%sTx %p]> style="filled" fillcolor=%q]`, r.key, write, r.tx, colorWritable[r.writeTx])
}
func (r *rbfQueryContext) dotClass() string {
return "rqc"
}
func (r *rbfQueryContext) dotId() string {
return fmt.Sprintf("rqc-%p", r)
}
func (r *rbfQueryContext) writeNode(w io.Writer) {
writeScope := "[read only]"
if r.scope != nil {
writeScope = r.scope.String()
if len(writeScope) > 20 {
writeScope = writeScope[:20] + "..."
}
writeScope = fmt.Sprintf("scope: %s", writeScope)
}
fmt.Fprintf(w, `[label=<rbfQueryContext<BR/>%s<BR/>%s> style="filled" fillcolor=%q]`, r.name, writeScope, colorWritable[r.scope != nil])
}
func (r *rbfQueryRead) dotClass() string {
return "qrw"
}
func (r *rbfQueryRead) dotId() string {
return fmt.Sprintf("qrw-%p", r)
}
func (r *rbfQueryRead) addTo(dg *dotGraph) {
dg.consider(r)
}
func (r *rbfQueryRead) writeNode(w io.Writer) {
fmt.Fprintf(w, `[label=<rbfQueryRead<BR/>[<FONT POINT-SIZE="12" FACE="courier">%s</FONT>]> style="filled" fillcolor=%q]`, r.fk, colorRead)
}
func (r *rbfQueryWrite) dotClass() string {
return "qrw"
}
func (r *rbfQueryWrite) dotId() string {
return fmt.Sprintf("qrw-%p", r)
}
func (r *rbfQueryWrite) addTo(dg *dotGraph) {
dg.consider(r)
}
func (r *rbfQueryWrite) writeNode(w io.Writer) {
fmt.Fprintf(w, `[label=<rbfQueryWrite<BR/>[<FONT POINT-SIZE="12" FACE="courier">%s</FONT>]> style="filled" fillcolor=%q]`, r.fk, colorWrite)
}
type unknownEntry struct {
data interface{}
}
func (u *unknownEntry) dotClass() string {
return "unk"
}
func (u *unknownEntry) dotId() string {
return fmt.Sprintf("unk-%p", u)
}
func (u *unknownEntry) writeNode(w io.Writer) {
fmt.Fprintf(w, "[label=\"unknown %T\"]", u.data)
}
func (u *unknownEntry) addTo(dg *dotGraph) {
dg.consider(u)
}