in_tail: Fix a stall bug on !follow_inode case (fluent#4327)

Fix fluent#3614

Although known stall issues of in_tail on `follow_inode` case are fixed
in v1.16.2, it has still a similar problem on `!follow_inode` case.

In this case, a tail watcher is possible to mark the position entry as
`unwatched` if it's tansitioned to `rotate_wait` state by
`refresh_watcher` even if another newer tail watcher is managing it.
It's hard to occur in usual because `stat_watcher` will be called
immediately after the file is changed  while `refresh_wather` is called
every 60 seconds by default. However, there is a rare possibility that
this order might be swapped especillay if in_tail is busy on processing
large amount of logs. Because in_tail is single threadied, event queues
such as timers or inotify will be stucked in this case.

There is no such problem on `follow_inode` case because position entries
are always marked as `unwatched` before entering `rotate_wait` state.

---------

Signed-off-by: Takuro Ashie <ashie@clear-code.com>
Co-authored-by: Daijiro Fukuda <fukuda@clear-code.com>

lib/fluent/plugin/in_tail.rb

@@ -564,7 +564,7 @@ def detach_watcher(tw, ino, close_io = true)
 
       tw.close if close_io
 
-      if tw.unwatched && @pf
+      if @pf && tw.unwatched && (@follow_inode || !@tails[tw.path])
         target_info = TargetInfo.new(tw.path, ino)
         @pf.unwatch(target_info)
       end

test/plugin/test_in_tail.rb

@@ -3017,4 +3017,109 @@ def test_path_resurrection
       )
     end
   end
+
+  sub_test_case "Update watchers for rotation without follow_inodes" do
+    # The scenario where in_tail wrongly unwatches the PositionEntry.
+    # This is reported in https://github.com/fluent/fluentd/issues/3614.
+    def test_refreshTW_during_rotation
+      config = config_element(
+        "ROOT",
+        "",
+        {
+          "path" => "#{@tmp_dir}/tail.txt0",
+          "pos_file" => "#{@tmp_dir}/tail.pos",
+          "tag" => "t1",
+          "format" => "none",
+          "read_from_head" => "true",
+          # In order to detach the old watcher quickly.
+          "rotate_wait" => "3s",
+          # In order to reproduce the same condition stably, ensure that `refresh_watchers` is not
+          # called by a timer.
+          "refresh_interval" => "1h",
+          # stat_watcher often calls `TailWatcher::on_notify` faster than creating a new log file,
+          # so disable it in order to reproduce the same condition stably.
+          "enable_stat_watcher" => "false",
+        }
+      )
+      d = create_driver(config, false)
+
+      tail_watchers = []
+      stub.proxy(d.instance).setup_watcher do |tw|
+        tail_watchers.append(tw)
+        tw
+      end
+
+      Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "wb") {|f| f.puts "file1 log1"}
+
+      d.run(expect_records: 6, timeout: 15) do
+        Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "ab") {|f| f.puts "file1 log2"}
+        FileUtils.move("#{@tmp_dir}/tail.txt0", "#{@tmp_dir}/tail.txt" + "1")
+
+        # This reproduces the following situation:
+        #     `refresh_watchers` is called during the rotation process and it detects the current file being lost.
+        #     Then it stops and unwatches the TailWatcher.
+        d.instance.refresh_watchers
+
+        Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "wb") {|f| f.puts "file2 log1"}
+
+        # `watch_timer` calls `TailWatcher::on_notify`, and then `update_watcher` trys to add the new TailWatcher.
+        # After `rotate_wait` interval, the PositionEntry is unwatched.
+        # HOWEVER, the new TailWatcher is still using that PositionEntry, so this breaks the PositionFile!!
+        # That PositionEntry is removed from `PositionFile::map`, but it is still working and remaining in the real pos file.
+        sleep 5
+
+        # Append to the new current log file.
+        # The PositionEntry is updated although it does not exist in `PositionFile::map`.
+        #     `PositionFile::map`: empty
+        #     Real pos file: `.../tail.txt 0000000000000016 (inode)`
+        Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "ab") {|f| f.puts "file2 log2"}
+
+        # Rotate again
+        [1, 0].each do |i|
+          FileUtils.move("#{@tmp_dir}/tail.txt#{i}", "#{@tmp_dir}/tail.txt#{i + 1}")
+        end
+        Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "wb") {|f| f.puts "file3 log1"}
+
+        # `watch_timer` calls `TailWatcher::on_notify`, and then `update_watcher` trys to update the TailWatcher.
+        sleep 3
+
+        Fluent::FileWrapper.open("#{@tmp_dir}/tail.txt0", "ab") {|f| f.puts "file3 log2"}
+
+        # Wait `rotate_wait` for file2 to make sure to close all IO handlers
+        sleep 3
+      end
+
+      inode_0 = tail_watchers[0]&.ino
+      inode_1 = tail_watchers[1]&.ino
+      inode_2 = tail_watchers[2]&.ino
+      record_values = d.events.collect { |event| event[2]["message"] }.sort
+      position_entries = []
+      Fluent::FileWrapper.open("#{@tmp_dir}/tail.pos", "r") do |f|
+        f.readlines(chomp: true).each do |line|
+          values = line.split("\t")
+          position_entries.append([values[0], values[1], values[2].to_i(16)])
+        end
+      end
+
+      assert_equal(
+        {
+          record_values: ["file1 log1", "file1 log2", "file2 log1", "file2 log2", "file3 log1", "file3 log2"],
+          tail_watcher_paths: ["#{@tmp_dir}/tail.txt0", "#{@tmp_dir}/tail.txt0", "#{@tmp_dir}/tail.txt0"],
+          tail_watcher_inodes: [inode_0, inode_1, inode_2],
+          tail_watcher_io_handler_opened_statuses: [false, false, false],
+          position_entries: [
+            # The recorded path is old, but it is no problem. The path is not used when using follow_inodes.
+            ["#{@tmp_dir}/tail.txt0", "0000000000000016", inode_2],
+          ],
+        },
+        {
+          record_values: record_values,
+          tail_watcher_paths: tail_watchers.collect { |tw| tw.path },
+          tail_watcher_inodes: tail_watchers.collect { |tw| tw.ino },
+          tail_watcher_io_handler_opened_statuses: tail_watchers.collect { |tw| tw.instance_variable_get(:@io_handler)&.opened? || false },
+          position_entries: position_entries
+        },
+      )
+    end
+  end
 end