X ΔTail draftly done

wcfs/todo.dot

@@ -77,6 +77,7 @@ digraph {
 	wcfsRead	[label="read(#blk)"]
 
 	blktabGet	[label="blktab.Get(#blk):\nmanually + → ⌈rev(#blk)⌉"]
+	δFtail		[style=filled fillcolor=grey95]
 
 	mappingRegister	[label="mmappings:\nregister/maint"]
 	clientInvHandle	[label="process\n#blk invalidations"]

wcfs/todo.svg

@@ -162,7 +162,7 @@
 <!-- δFtail -->
 <g id="node10" class="node">
 <title>δFtail</title>
-<ellipse fill="none" stroke="#000000" cx="797.8112" cy="-188.6102" rx="31.6951" ry="18"/>
+<ellipse fill="#f2f2f2" stroke="#000000" cx="797.8112" cy="-188.6102" rx="31.6951" ry="18"/>
 <text text-anchor="middle" x="797.8112" y="-184.9102" font-family="Times,serif" font-size="14.00" fill="#000000">δFtail</text>
 </g>
 <!-- wcfsInvProcess&#45;&gt;δFtail -->

wcfs/wcfs.go

@@ -268,7 +268,7 @@ package main
 //	- if retrieved successfully -> store retrieved data back into OS file
 //	  cache for file/@<rev>/data[blk], where
 //
-//	    rev = max(δFtail.by(#blk)) || zconn.at	; see below about δFtail
+//	    rev = max(δFtail.by(#blk)) || min(rev ∈ δFtail) || zconn.at	; see below about δFtail
 //
 //	- invalidate file/head/data[blk] in OS file cache.
 //
@@ -279,7 +279,7 @@ package main
 //
 // 5) for every file δFtail invalidation info about head/data is maintained:
 //
-//	- tail: of [](rev↑, []#blk)
+//	- tailv: [](rev↑, []#blk)
 //	- by:    {} #blk -> []rev↑ in tail
 //
 //    δFtail.tail describes invalidations to file we learned from ZODB invalidation.
@@ -287,8 +287,6 @@ package main
 //
 //    min(rev) in δFtail is min(@at) at which head/data is currently mmapped (see below).
 //
-//    TODO invariant -> δFtail - which entries are present?
-//
 // 6) when we receive a FUSE read(#blk) request to a file/head/data we process it as follows:
 //
 //   6.1) load blkdata for head/data[blk] @zconn.at .
@@ -307,7 +305,7 @@ package main
 //
 //	or another upper bound if #blk ∉ δFtail:
 //
-//	  rev(blk) ≤ min(rev) in δFtail		; #blk ∉ δFtail
+//	  rev(blk) ≤ min(rev ∈ δFtail)		; #blk ∉ δFtail
 //
 //
 //	below rev'(blk) is min(of the estimates found):
@@ -320,7 +318,7 @@ package main
 //	- rev'(blk) ≤ at: -> do nothing		XXX || blk ∉ mapping
 //	- rev'(blk) > at:
 //	  - if blk ∈ mmapping.pinned -> do nothing
-//	  - rev = max(δFtail.by(#blk) : _ ≤ at)	|| at
+//	  - rev = max(δFtail.by(#blk) : _ ≤ at)	|| min(rev ∈ δFtail : rev ≤ at)	|| at
 //	  - client.remmap(addr[blk], file/@rev/data)
 //	  - mmapping.pinned += blk
 //

wcfs/δtail.go

@@ -20,19 +20,21 @@
 // XXX -> internal/δtail/ ?
 
 package main
-// δtail maintenance	XXX
+// δtail maintenance
 
 import (
+	"fmt"
+
 	"lab.nexedi.com/kirr/neo/go/zodb"
 )
 
-type ID int64	// XXX -> template
+type ID int64	// XXX -> template ?
 
 // ΔTail represents tail of revisional changes.
 //
 // It semantically consists of
 //
-//	[] of (rev↑, []id)
+//	[](rev↑, []id)
 //
 // where
 //
@@ -46,7 +48,7 @@ type ID int64	// XXX -> template
 //	- query the tail about what is last revision that changed an id.
 //
 // It is generally not safe to use ΔTail from multiple goroutines simultaneously.
-// It is safe to perform multiple simultaneous queries.
+// It is safe to perform multiple simultaneous read-kind operations.
 //
 // (*) examples of id:
 //
@@ -54,16 +56,16 @@ type ID int64	// XXX -> template
 //	#blk - file block number, when ΔTail represents changes to a file.
 type ΔTail struct {
 	tailv []δRevEntry
-	lastRevOf map[ID]zodb.Tid
+
+	lastRevOf map[ID]zodb.Tid // index for LastRevOf queries
 	// TODO also add either tailv idx <-> rev index, or lastRevOf -> tailv idx
 	// (if linear back-scan of δRevEntry starts eat cpu).
 }
 
 // δRevEntry represents information of what have been changed in one revision.
-// XXX naming
 type δRevEntry struct {
 	rev     zodb.Tid
-	δv      []ID
+	changev []ID
 }
 
 // NewΔTail creates new ΔTail object.
@@ -75,10 +77,17 @@ func NewΔTail() *ΔTail {
 // Append appends to δtail information about what have been changed in next revision.
 //
 // rev must be ↑.
-func (δtail *ΔTail) Append(rev zodb.Tid, δv []ID) {
-	// XXX check rev↑
-	δtail.tailv = append(δtail.tailv, δRevEntry{rev, δv})
-	for _, id := range δv {
+func (δtail *ΔTail) Append(rev zodb.Tid, changev []ID) {
+	// check rev↑
+	// XXX better also check even when δtail is ø (after forget)
+	if l := len(δtail.tailv); l > 0 {
+		if revPrev := δtail.tailv[l-1].rev; revPrev >= rev {
+			panic(fmt.Sprintf("δtail.Append: rev not ↑: %s -> %s", revPrev, rev))
+		}
+	}
+
+	δtail.tailv = append(δtail.tailv, δRevEntry{rev, changev})
+	for _, id := range changev {
 		δtail.lastRevOf[id] = rev
 	}
 }
@@ -93,51 +102,86 @@ func (δtail *ΔTail) ForgetBefore(revCut zodb.Tid) {
 		}
 		icut = i+1
 
-		// if forgottent revision was last for id, we have to update lastRevOf index
-		for _, id := range δ.δv {
+		// if forgotten revision was last for id, we have to update lastRevOf index
+		for _, id := range δ.changev {
 			if δtail.lastRevOf[id] == rev {
 				delete(δtail.lastRevOf, id)
 			}
 		}
 	}
 
-	// tailv = tailv[icut:]	  but without growing underlying storage array indefinetely
-	copy(δtail.tailv, δtail.tailv[icut:])
-	δtail.tailv = δtail.tailv[:len(δtail.tailv)-icut]
+	// tailv = tailv[icut:]	  but without
+	// 1) growing underlying storage array indefinitely
+	// 2) keeping underlying storage after forget
+	l := len(δtail.tailv)-icut
+	tailv := make([]δRevEntry, l)
+	copy(tailv, δtail.tailv[icut:])
+	δtail.tailv = tailv
 }
 
-// LastRevOf returns what was the last revision that changed id as of at database state.
+// LastRevOf tries to return what was the last revision that changed id as of at database state.
+//
+// Depending on current information in δtail it returns either exact result, or
+// an upper-bound estimate for the last id revision, assuming id was changed ≤ at:
+//
+// 1) if δtail does not cover at, at is returned:
 //
-// in other words
+//	# at ∉ [min(rev ∈ δtail), max(rev ∈ δtail)]
+//	LastRevOf(id, at) = at
 //
+// 2) if δtail has an entry corresponding to id change, it gives exactly the
+//    last revision that changed id:
+//
+//	# at ∈ [min(rev ∈ δtail), max(rev ∈ δtail)]
+//	# ∃ rev ∈ δtail: rev changed id && rev ≤ at
 //	LastRevOf(id, at) = max(rev: rev changed id && rev ≤ at)
 //
-// XXX if δtail does not contain records with id -> what is returned?
-func (δtail *ΔTail) LastRevOf(id ID, at zodb.Tid) zodb.Tid {
+// 3) if δtail does not contain appropriate record with id - it returns δtail's
+//    lower bound as the estimate for the upper bound of the last id revision:
+//
+//	# at ∈ [min(rev ∈ δtail), max(rev ∈ δtail)]
+//	# ∄ rev ∈ δtail: rev changed id && rev ≤ at
+//	LastRevOf(id, at) = min(rev ∈ δtail)
+//
+// On return exact indicates whether returned revision is exactly the last
+// revision of id, or only an upper-bound estimate of it.
+func (δtail *ΔTail) LastRevOf(id ID, at zodb.Tid) (_ zodb.Tid, exact bool) {
+	// check if we have no coverage at all
+	l := len(δtail.tailv)
+	if l == 0 {
+		return at, false
+	}
+	revMin := δtail.tailv[0].rev
+	revMax := δtail.tailv[l-1].rev
+	if !(revMin <= at && at <= revMax) {
+		return at, false
+	}
+
+	// we have the coverage
 	rev, ok := δtail.lastRevOf[id]
 	if !ok {
-		panic("TODO")	// XXX
+		return δtail.tailv[0].rev, false
 	}
 
 	if rev <= at {
-		return rev
+		return rev, true
 	}
 
 	// what's in index is after at - scan tailv back to find appropriate entry
 	// XXX linear scan
-	for i := len(δtail.tailv) - 1; i >= 0; i-- {
+	for i := l - 1; i >= 0; i-- {
 		δ := δtail.tailv[i]
 		if δ.rev > at {
 			continue
 		}
 
-		for _, δid := range δ.δv {
+		for _, δid := range δ.changev {
 			if id == δid {
-				return δ.rev
+				return δ.rev, true
 			}
 		}
 	}
 
 	// nothing found
-	panic("TODO")	// XXX
+	return δtail.tailv[0].rev, false
 }

wcfs/δtail_test.go

@@ -20,6 +20,7 @@
 package main
 
 import (
+	"fmt"
 	"reflect"
 	"testing"
 
@@ -33,14 +34,14 @@ func TestΔTail(t *testing.T) {
 
 	δtail := NewΔTail()
 
-	// R is syntic sugar to create 1 δRevEntry
-	R := func(rev zodb.Tid, δv ...ID) δRevEntry {
-		return δRevEntry{rev, δv}
+	// R is syntactic sugar to create 1 δRevEntry
+	R := func(rev zodb.Tid, changev ...ID) δRevEntry {
+		return δRevEntry{rev, changev}
 	}
 
-	// δAppend is syntatic sugar for δtail.Append
+	// δAppend is syntactic sugar for δtail.Append
 	δAppend := func(δ δRevEntry) {
-		δtail.Append(δ.rev, δ.δv)
+		δtail.Append(δ.rev, δ.changev)
 	}
 
 	// δCheck verifies that δtail state corresponds to provided tailv
@@ -60,10 +61,10 @@ func TestΔTail(t *testing.T) {
 		// verify lastRevOf query / index
 		lastRevOf := make(map[ID]zodb.Tid)
 		for _, δ := range tailv {
-			for _, id := range δ.δv {
-				idRev := δtail.LastRevOf(id, δ.rev)
-				if idRev != δ.rev {
-					t.Fatalf("LastRevOf(%v, at=%s) -> %s  ; want %s", id, δ.rev, idRev, δ.rev)
+			for _, id := range δ.changev {
+				idRev, exact := δtail.LastRevOf(id, δ.rev)
+				if !(idRev == δ.rev && exact) {
+					t.Fatalf("LastRevOf(%v, at=%s) -> %s, %v  ; want %s, %v", id, δ.rev, idRev, exact, δ.rev, true)
 				}
 
 				lastRevOf[id] = δ.rev
@@ -76,37 +77,87 @@ func TestΔTail(t *testing.T) {
 
 	}
 
+	// δCheckLastUP verifies that δtail.LastRevOf(id, at) gives lastOk and exact=false.
+	// (we don't need to check for exact=true as those cases are covered in δCheck)
+	δCheckLastUP := func(id ID, at, lastOk zodb.Tid) {
+		t.Helper()
+
+		last, exact := δtail.LastRevOf(id, at)
+		if !(last == lastOk && exact == false) {
+			t.Fatalf("LastRevOf(%v, at=%s) -> %s, %v  ; want %s, %v", id, at, last, exact, lastOk, false)
+		}
+	}
+
 
 	δCheck()
+	δCheckLastUP(4, 12, 12)	// δtail = ø
 
 	δAppend(R(10, 3,5))
 	δCheck(R(10, 3,5))
 
+	δCheckLastUP(3,  9,  9)	// at < δtail
+	δCheckLastUP(3, 12, 12)	// at > δtail
+	δCheckLastUP(4, 10, 10)	// id ∉ δtail
+
 	δAppend(R(11, 7))
 	δCheck(R(10, 3,5), R(11, 7))
 
 	δAppend(R(12, 7))
 	δCheck(R(10, 3,5), R(11, 7), R(12, 7))
 
-	δAppend(R(14, 3,7))
-	δCheck(R(10, 3,5), R(11, 7), R(12, 7), R(14, 3,7))
+	δAppend(R(14, 3,8))
+	δCheck(R(10, 3,5), R(11, 7), R(12, 7), R(14, 3,8))
+
+	δCheckLastUP(8, 12, 10) // id ∈ δtail, but has no entry with rev ≤ at
 
 	δtail.ForgetBefore(10)
-	δCheck(R(10, 3,5), R(11, 7), R(12, 7), R(14, 3,7))
+	δCheck(R(10, 3,5), R(11, 7), R(12, 7), R(14, 3,8))
 
 	δtail.ForgetBefore(11)
-	δCheck(R(11, 7), R(12, 7), R(14, 3,7))
+	δCheck(R(11, 7), R(12, 7), R(14, 3,8))
 
 	δtail.ForgetBefore(13)
-	δCheck(R(14, 3,7))
+	δCheck(R(14, 3,8))
 
 	δtail.ForgetBefore(15)
 	δCheck()
 
+	// Append panics on non-↑ rev
+	δAppend(R(15, 1))
+	func() {
+		defer func() {
+			r := recover()
+			if r == nil {
+				t.Fatal("append non-↑: not panicked")
+			}
+			rev := zodb.Tid(15)
+			want := fmt.Sprintf("δtail.Append: rev not ↑: %s -> %s", rev, rev)
+			if r != want {
+				t.Fatalf("append non-↑:\nhave: %q\nwant: %q", r, want)
+			}
+		}()
+
+		δAppend(R(15, 1))
+	}()
+
+
+	// .tailv underlying storage is not kept after forget
+	δtail.ForgetBefore(16)
+
+	const N = 1E3
+	for rev, i := zodb.Tid(16), 0; i < N; i, rev = i+1, rev+1 {
+		δAppend(R(rev, 1))
+	}
+
+	capN := cap(δtail.tailv)
+	δtail.ForgetBefore(N)
+	if c := cap(δtail.tailv); !(c < capN/10) {
+		t.Fatalf("forget: tailv storage did not shrink: cap%v: %d -> cap: %d", N, capN, c)
+	}
 
-	// XXX .Append(rev not ↑ - panic)
-	// XXX edge cases
-	// XXX .tailv underlying storage does not grow indefinitely
+	// .tailv underlying storage does not grow indefinitely
+	// XXX cannot test as the growth here goes to left and we cannot get
+	// access to whole underlying array from a slice.
 }
 
 func tailvEqual(a, b []δRevEntry) bool {