go/zodb/fs1: Teach it to support notifications on database change

Following-up on 4d2c8b1d (go/zodb: Require drivers to provide
notifications for database change events) let's teach FileStorage to
support watching for database changes:

- add watcher who observes changes to data file via fsnotify.

- when we see that the file was updated - there is a tricky case to distinguish
  in-progress writes from just some garbage added at tail. See comments in
  _watcher for details.

- if the watcher sees there is indeed new transactions, the watcher
  updates index & txnh{Min,Max} and sends corresponding event to watchq.

- since index / txnh{Min,Max} can now be changed (by watcher) -> they
  are now protected by mu.

- consequently operations like LastTid, LastOid, Load, ... are all taught to
  observe index / txnh{Min,Max} state in read-snapshot way.

- add down & friends to indicate that the storage is no longer
  operational, if watcher sees there is a problem with data file.

- Open is reworked to start from loading/rebuilding index first, and then tail
  to watcher to detect whether what's after found topPos is garbage or another
  in-progress transaction. Consequently it is now possible to correctly open
  filestorage that is being currently written to and has in-progress transaction
  at tail.

The patch is a bit big, but the changes here are all tightly
interrelated.

go/zodb/storage/fs1/filestorage_test.go

-// Copyright (C) 2017  Nexedi SA and Contributors.
-//                     Kirill Smelkov <kirr@nexedi.com>
+// Copyright (C) 2017-2019  Nexedi SA and Contributors.
+//                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
 // it under the terms of the GNU General Public License version 3, or (at your
@@ -20,15 +20,21 @@
 package fs1
 
 import (
+	"bytes"
 	"context"
 	"fmt"
 	"io"
+	"io/ioutil"
+	"os"
+	"os/exec"
 	"reflect"
 	"testing"
 
+	"lab.nexedi.com/kirr/neo/go/internal/xtesting"
 	"lab.nexedi.com/kirr/neo/go/zodb"
 
 	"lab.nexedi.com/kirr/go123/exc"
+	"lab.nexedi.com/kirr/go123/xerr"
 )
 
 // one database transaction record
@@ -109,7 +115,13 @@ func checkLoad(t *testing.T, fs *FileStorage, xid zodb.Xid, expect objState) {
 }
 
 func xfsopen(t testing.TB, path string) *FileStorage {
-	fs, err := Open(context.Background(), path)
+	t.Helper()
+	return xfsopenopt(t, path, &zodb.DriverOptions{ReadOnly: true})
+}
+
+func xfsopenopt(t testing.TB, path string, opt *zodb.DriverOptions) *FileStorage {
+	t.Helper()
+	fs, err := Open(context.Background(), path, opt)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -117,7 +129,7 @@ func xfsopen(t testing.TB, path string) *FileStorage {
 }
 
 func TestLoad(t *testing.T) {
-	fs := xfsopen(t, "testdata/1.fs") // TODO open read-only
+	fs := xfsopen(t, "testdata/1.fs")
 	defer exc.XRun(fs.Close)
 
 	// current knowledge of what was "before" for an oid as we scan over
@@ -266,7 +278,7 @@ func testIterate(t *testing.T, fs *FileStorage, tidMin, tidMax zodb.Tid, expectv
 }
 
 func TestIterate(t *testing.T) {
-	fs := xfsopen(t, "testdata/1.fs") // TODO open ro
+	fs := xfsopen(t, "testdata/1.fs")
 	defer exc.XRun(fs.Close)
 
 	// all []tids in test database
@@ -302,7 +314,7 @@ func TestIterate(t *testing.T) {
 }
 
 func BenchmarkIterate(b *testing.B) {
-	fs := xfsopen(b, "testdata/1.fs") // TODO open ro
+	fs := xfsopen(b, "testdata/1.fs")
 	defer exc.XRun(fs.Close)
 
 	ctx := context.Background()
@@ -341,3 +353,188 @@ func BenchmarkIterate(b *testing.B) {
 
 	b.StopTimer()
 }
+
+// TestWatch verifies that watcher can observe commits done from outside.
+func TestWatch(t *testing.T) {
+	X := exc.Raiseif
+
+	xtesting.NeedPy(t, "zodbtools")
+	workdir := xworkdir(t)
+	tfs := workdir + "/t.fs"
+
+	// Object represents object state to be committed.
+	type Object struct {
+		oid  zodb.Oid
+		data string
+	}
+
+	// zcommit commits new transaction into tfs with data specified by objv.
+	zcommit := func(at zodb.Tid, objv ...Object) (_ zodb.Tid, err error) {
+		defer xerr.Contextf(&err, "zcommit @%s", at)
+
+		// prepare text input for `zodb commit`
+		zin := &bytes.Buffer{}
+		fmt.Fprintf(zin, "user %q\n", "author")
+		fmt.Fprintf(zin, "description %q\n", fmt.Sprintf("test commit; at=%s", at))
+		fmt.Fprintf(zin, "extension %q\n", "")
+		for _, obj := range objv {
+			fmt.Fprintf(zin, "obj %s %d null:00\n", obj.oid, len(obj.data))
+			zin.WriteString(obj.data)
+			zin.WriteString("\n")
+		}
+		zin.WriteString("\n")
+
+		// run py `zodb commit`
+		cmd:= exec.Command("python2", "-m", "zodbtools.zodb", "commit", tfs, at.String())
+		cmd.Stdin  = zin
+		cmd.Stderr = os.Stderr
+		out, err := cmd.Output()
+		if err != nil {
+			return zodb.InvalidTid, err
+		}
+
+		out = bytes.TrimSuffix(out, []byte("\n"))
+		tid, err := zodb.ParseTid(string(out))
+		if err != nil {
+			return zodb.InvalidTid, fmt.Errorf("committed, but invalid output: %s", err)
+		}
+
+		return tid, nil
+	}
+
+	xcommit := func(at zodb.Tid, objv ...Object) zodb.Tid {
+		t.Helper()
+		tid, err := zcommit(at, objv...); X(err)
+		return tid
+	}
+
+	// force tfs creation & open tfs at go side
+	at := xcommit(0, Object{0, "data0"})
+
+	watchq := make(chan zodb.CommitEvent)
+	fs := xfsopenopt(t, tfs, &zodb.DriverOptions{ReadOnly: true, Watchq: watchq})
+	ctx := context.Background()
+
+	checkLastTid := func(lastOk zodb.Tid) {
+		t.Helper()
+		head, err := fs.LastTid(ctx); X(err)
+		if head != lastOk {
+			t.Fatalf("check last_tid: got %s;  want %s", head, lastOk)
+		}
+	}
+
+	checkLastTid(at)
+
+	checkLoad := func(at zodb.Tid, oid zodb.Oid, dataOk string, serialOk zodb.Tid) {
+		t.Helper()
+		xid := zodb.Xid{at, oid}
+		buf, serial, err := fs.Load(ctx, xid); X(err)
+		data := string(buf.XData())
+		if !(data == dataOk && serial == serialOk) {
+			t.Fatalf("check load %s:\nhave: %q %s\nwant: %q %s",
+				xid, data, serial, dataOk, serialOk)
+		}
+	}
+
+	// commit -> check watcher observes what we committed.
+	//
+	// XXX python `import pkg_resources` takes ~ 300ms.
+	// https://github.com/pypa/setuptools/issues/510
+	//
+	// Since pkg_resources are used everywhere (e.g. in zodburi to find all
+	// uri resolvers) this import slowness becomes the major part of time to
+	// run py `zodb commit`.
+	//
+	// if one day it is either fixed, or worked around, we could ↑ 10 to 100.
+	for i := zodb.Oid(1); i <= 10; i++ {
+		data0 := fmt.Sprintf("data0.%d", i)
+		datai := fmt.Sprintf("data%d", i)
+		at = xcommit(at,
+			Object{0, data0},
+			Object{i, datai})
+
+		// TODO also test for watcher errors
+		e := <-watchq
+
+		if objvWant := []zodb.Oid{0, i}; !(e.Tid == at && reflect.DeepEqual(e.Changev, objvWant)) {
+			t.Fatalf("watch:\nhave: %s %s\nwant: %s %s", e.Tid, e.Changev, at, objvWant)
+		}
+
+		checkLastTid(at)
+
+		// make sure we can load what was committed.
+		checkLoad(at, 0, data0, at)
+		checkLoad(at, i, datai, at)
+	}
+
+	err := fs.Close(); X(err)
+
+	e, ok := <-watchq
+	if ok {
+		t.Fatalf("watch after close -> %v;  want closed", e)
+	}
+}
+
+// TestOpenRecovery verifies how Open handles data file with not-finished voted
+// transaction in the end.
+func TestOpenRecovery(t *testing.T) {
+	X := exc.Raiseif
+	main, err := ioutil.ReadFile("testdata/1.fs"); X(err)
+	index, err := ioutil.ReadFile("testdata/1.fs.index"); X(err)
+	lastTidOk := _1fs_dbEntryv[len(_1fs_dbEntryv)-1].Header.Tid
+	topPos := int64(_1fs_indexTopPos)
+	voteTail, err := ioutil.ReadFile("testdata/1voted.tail"); X(err)
+
+	workdir := xworkdir(t)
+	ctx := context.Background()
+
+	// checkL runs f on main + voteTail[:l] . Two cases are verified:
+	// 1) when index is already present, and
+	// 2) when initially there is no index.
+	checkL := func(t *testing.T, l int, f func(t *testing.T, tfs string)) {
+		tfs := fmt.Sprintf("%s/1+vote%d.fs", workdir, l)
+		t.Run(fmt.Sprintf("oldindex=n/tail=+vote%d", l), func(t *testing.T) {
+			err := ioutil.WriteFile(tfs, append(main, voteTail[:l]...), 0600); X(err)
+			f(t, tfs)
+		})
+		t.Run(fmt.Sprintf("oldindex=y/tail=+vote%d", l), func(t *testing.T) {
+			err := ioutil.WriteFile(tfs+".index", index, 0600); X(err)
+			f(t, tfs)
+		})
+	}
+
+	// if txn header can be fully read - it should be all ok
+	lok := []int{0}
+	for l := len(voteTail); l >= TxnHeaderFixSize; l-- {
+		lok = append(lok, l)
+	}
+	for _, l := range lok {
+		checkL(t, l, func(t *testing.T, tfs string) {
+			fs := xfsopen(t, tfs)
+			defer func() {
+				err = fs.Close(); X(err)
+			}()
+			head, err := fs.LastTid(ctx); X(err)
+			if head != lastTidOk {
+				t.Fatalf("last_tid: %s  ; expected: %s", head, lastTidOk)
+			}
+		})
+	}
+
+	// if txn header is stably incomplete - open should fail
+	// XXX better check 0..sizeof(txnh)-1 but in this range each check is slow.
+	for _, l := range []int{TxnHeaderFixSize-1,1} {
+		checkL(t, l, func(t *testing.T, tfs string) {
+			_, err := Open(ctx, tfs, &zodb.DriverOptions{ReadOnly: true})
+			estr := ""
+			if err != nil {
+				estr = err.Error()
+			}
+			ewant := fmt.Sprintf("open %s: checking whether it is garbage @%d: %s",
+				tfs, topPos, &RecordError{tfs, "transaction record", topPos, "read", io.ErrUnexpectedEOF})
+			if estr != ewant {
+				t.Fatalf("unexpected error:\nhave: %q\nwant: %q", estr, ewant)
+			}
+		})
+	}
+}

go/zodb/storage/fs1/format.go

-// Copyright (C) 2017-2018  Nexedi SA and Contributors.
+// Copyright (C) 2017-2019  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -152,7 +152,7 @@ func (fh *FileHeader) Load(r io.ReaderAt) error {
 	_, err := r.ReadAt(fh.Magic[:], 0)
 	err = okEOF(err)
 	if err != nil {
-		return err
+		return fmt.Errorf("%sread magic: %s", ioprefix(r), err)
 	}
 	if string(fh.Magic[:]) != Magic {
 		return fmt.Errorf("%sinvalid fs1 magic %q", ioprefix(r), fh.Magic)

go/zodb/storage/fs1/index.go

-// Copyright (C) 2017-2018  Nexedi SA and Contributors.
+// Copyright (C) 2017-2019  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -453,7 +453,10 @@ func (index *Index) Update(ctx context.Context, r io.ReaderAt, topPos int64, pro
 			return err
 		}
 
-		// XXX check txnh.Status != TxnInprogress
+		// this transaction was only voted, not fully committed yet.
+		if it.Txnh.Status == zodb.TxnInprogress {
+			return nil
+		}
 
 		// check for topPos overlapping txn & whether we are done.
 		// topPos=-1 will never match here
@@ -469,7 +472,9 @@ func (index *Index) Update(ctx context.Context, r io.ReaderAt, topPos int64, pro
 		// do not update the index immediately so that in case of error
 		// in the middle of txn's data, index stays consistent and
 		// correct for topPos pointing to previous transaction.
-		update := map[zodb.Oid]int64{} // XXX malloc every time -> better reuse
+		//
+		// (keep in sync with FileStorage.watcher)
+		update := map[zodb.Oid]int64{}
 		for {
 			err = it.NextData()
 			if err != nil {
@@ -693,7 +698,9 @@ func (index *Index) VerifyForFile(ctx context.Context, path string, ntxn int, pr
 		return nil, err
 	}
 
-	topPos := fi.Size() // XXX there might be last TxnInprogress transaction
+	// FIXME there might be last TxnInprogress transaction.
+	// TODO  -> try to read txn header, and if it is ø or in-progress - that's ok.
+	topPos := fi.Size()
 	if index.TopPos != topPos {
 		return nil, indexCorrupt(f, "topPos mismatch: data=%v  index=%v", topPos, index.TopPos)
 	}

go/zodb/storage/fs1/py/gen-testdata

 #!/usr/bin/env python2
 # -*- coding: utf-8 -*-
-# Copyright (C) 2017-2018  Nexedi SA and Contributors.
+# Copyright (C) 2017-2019  Nexedi SA and Contributors.
 #                          Kirill Smelkov <kirr@nexedi.com>
 #
 # This program is free software: you can Use, Study, Modify and Redistribute
@@ -21,7 +21,10 @@
 """generate reference fs1 database and index for tests"""
 
 from ZODB.FileStorage import FileStorage
-from zodbtools.test.gen_testdata import gen_testdb
+from ZODB import DB
+from zodbtools.test.gen_testdata import gen_testdb, precommit
+from os import stat, remove
+from shutil import copyfile
 from golang.gcompat import qq
 
 import struct
@@ -133,6 +136,38 @@ def main():
             emit("\t\t},")
             emit("\t},")
         emit("}")
+    stor.close()
+
+
+    # prepare file with voted (not fully committed) tail
+    voted = "testdata/1voted.fs"
+    copyfile(outfs, voted)
+    vstor = FileStorage(voted)
+    vdb   = DB(vstor)
+    vconn = vdb.open()
+    vroot = vconn.root()
+    vroot._p_activate() # to know its current serial
+
+    txn = precommit(u"author", u"description", {'aaa': 'bbb'})
+    vstor.tpc_begin(txn)
+    vstor.store(vroot._p_oid, vroot._p_serial, '000 data 000', '', txn)
+    vstor.tpc_vote(txn)
+    # NO tpc_finish here so that status remain 'c' (voted) instead of ' ' (committed)
+
+    st = stat(outfs)
+    l  = st.st_size
+    vf = open(voted, 'rb')
+    vf.seek(l)
+    voted_tail = vf.read()
+    assert voted_tail[-1+8+8+1] == 'c'  # voted, not finished (' ')
+
+    with open("testdata/1voted.tail", "wb") as vt:
+        vt.write(voted_tail)
+
+    remove(voted)
+    remove(voted+".index")
+    remove(voted+".tmp")
+    remove(voted+".lock")
 
 
 if __name__ == '__main__':

go/zodb/storage/fs1/url.go

@@ -22,8 +22,6 @@ package fs1
 
 import (
 	"context"
-	"fmt"
-	"log"
 	"net/url"
 
 	"lab.nexedi.com/kirr/neo/go/zodb"
@@ -34,21 +32,7 @@ func openByURL(ctx context.Context, u *url.URL, opt *zodb.DriverOptions) (zodb.I
 	// XXX u.Path is not always raw path - recheck and fix
 	path := u.Host + u.Path
 
-	// XXX readonly stub
-	// XXX place = ?
-	if !opt.ReadOnly {
-		return nil, fmt.Errorf("fs1: %s: TODO write mode not implemented", path)
-	}
-
-	// FIXME handle opt.Watchq
-	// for now we pretend as if the database is not changing.
-	if opt.Watchq != nil {
-		log.Print("fs1: FIXME: watchq support not implemented - there" +
-			  "won't be notifications about database changes")
-	}
-
-	fs, err := Open(ctx, path)
-	fs.watchq = opt.Watchq
+	fs, err := Open(ctx, path, opt)
 	return fs, err
 }