Make sure view compaction terminates

If a view group is compacting and the corresponding database
is shutdown by the LRU system, then the view compaction is
aborted because its couch view group process shutdowns.
This could lead to situations where the number of active
databases is much higher than max_dbs_open and making it
impossible to compact view groups.

Issue reported and patch tested by Mike Leddy. Thanks.

COUCHDB-1283

This is a backport of revision 1171328 from branch 1.2.x



git-svn-id: https://svn.apache.org/repos/asf/couchdb/branches/1.1.x@1171329 13f79535-47bb-0310-9956-ffa450edef68

3 files changed, 302 insertions, 1 deletions

diff --git a/test/etap/200-view-group-no-db-leaks.t b/test/etap/200-view-group-no-db-leaks.t
index 9c77f1a8..f506b7dc 100755
--- a/test/etap/200-view-group-no-db-leaks.t
+++ b/test/etap/200-view-group-no-db-leaks.t
@@ -165,7 +165,7 @@ wait_db_compact_done(N) ->
     end.
 
 compact_view_group() ->
-    ok = couch_view_compactor:start_compact(test_db_name(), ddoc_name()),
+    {ok, _} = couch_view_compactor:start_compact(test_db_name(), ddoc_name()),
     wait_view_compact_done(10).
 
 wait_view_compact_done(0) ->
diff --git a/test/etap/201-view-group-shutdown.t b/test/etap/201-view-group-shutdown.t
new file mode 100755
index 00000000..03feac2b
--- /dev/null
+++ b/test/etap/201-view-group-shutdown.t
@@ -0,0 +1,300 @@
+#!/usr/bin/env escript
+%% -*- erlang -*-
+
+% Licensed under the Apache License, Version 2.0 (the "License"); you may not
+% use this file except in compliance with the License. You may obtain a copy of
+% the License at
+%
+%   http://www.apache.org/licenses/LICENSE-2.0
+%
+% Unless required by applicable law or agreed to in writing, software
+% distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+% WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+% License for the specific language governing permissions and limitations under
+% the License.
+
+-record(user_ctx, {
+    name = null,
+    roles = [],
+    handler
+}).
+
+-record(db, {
+    main_pid = nil,
+    update_pid = nil,
+    compactor_pid = nil,
+    instance_start_time, % number of microsecs since jan 1 1970 as a binary string
+    fd,
+    fd_ref_counter,
+    header = nil,
+    committed_update_seq,
+    fulldocinfo_by_id_btree,
+    docinfo_by_seq_btree,
+    local_docs_btree,
+    update_seq,
+    name,
+    filepath,
+    validate_doc_funs = [],
+    security = [],
+    security_ptr = nil,
+    user_ctx = #user_ctx{},
+    waiting_delayed_commit = nil,
+    revs_limit = 1000,
+    fsync_options = [],
+    is_sys_db = false
+}).
+
+main_db_name() -> <<"couch_test_view_group_shutdown">>.
+
+
+main(_) ->
+    test_util:init_code_path(),
+
+    etap:plan(17),
+    case (catch test()) of
+        ok ->
+            etap:end_tests();
+        Other ->
+            etap:diag(io_lib:format("Test died abnormally: ~p", [Other])),
+            etap:bail(Other)
+    end,
+    ok.
+
+
+test() ->
+    couch_server_sup:start_link(test_util:config_files()),
+    ok = couch_config:set("couchdb", "max_dbs_open", "3", false),
+    ok = couch_config:set("couchdb", "delayed_commits", "false", false),
+    crypto:start(),
+
+    % Test that while a view group is being compacted its database can not
+    % be closed by the database LRU system.
+    test_view_group_compaction(),
+
+    couch_server_sup:stop(),
+    ok.
+
+
+test_view_group_compaction() ->
+    {ok, DbWriter3} = create_db(<<"couch_test_view_group_shutdown_w3">>),
+    ok = couch_db:close(DbWriter3),
+
+    {ok, MainDb} = create_main_db(),
+    ok = couch_db:close(MainDb),
+
+    {ok, DbWriter1} = create_db(<<"couch_test_view_group_shutdown_w1">>),
+    ok = couch_db:close(DbWriter1),
+
+    {ok, DbWriter2} = create_db(<<"couch_test_view_group_shutdown_w2">>),
+    ok = couch_db:close(DbWriter2),
+
+    Writer1 = spawn_writer(DbWriter1#db.name),
+    Writer2 = spawn_writer(DbWriter2#db.name),
+    etap:is(is_process_alive(Writer1), true, "Spawned writer 1"),
+    etap:is(is_process_alive(Writer2), true, "Spawned writer 2"),
+
+    etap:is(get_writer_status(Writer1), ok, "Writer 1 opened his database"),
+    etap:is(get_writer_status(Writer2), ok, "Writer 2 opened his database"),
+
+    {ok, CompactPid} = couch_view_compactor:start_compact(
+        MainDb#db.name, <<"foo">>),
+    MonRef = erlang:monitor(process, CompactPid),
+
+    % Add some more docs to database and trigger view update
+    {ok, MainDb2} = couch_db:open_int(MainDb#db.name, []),
+    ok = populate_main_db(MainDb2, 3, 3),
+    update_view(MainDb2#db.name, <<"_design/foo">>, <<"foo">>),
+    ok = couch_db:close(MainDb2),
+
+    % Assuming the view compaction takes more than 50ms to complete
+    ok = timer:sleep(50),
+    Writer3 = spawn_writer(DbWriter3#db.name),
+    etap:is(is_process_alive(Writer3), true, "Spawned writer 3"),
+
+    etap:is(get_writer_status(Writer3), {error, all_dbs_active},
+        "Writer 3 got {error, all_dbs_active} when opening his database"),
+
+    etap:is(is_process_alive(Writer1), true, "Writer 1 still alive"),
+    etap:is(is_process_alive(Writer2), true, "Writer 2 still alive"),
+    etap:is(is_process_alive(Writer3), true, "Writer 3 still alive"),
+
+    receive
+    {'DOWN', MonRef, process, CompactPid, normal} ->
+         etap:diag("View group compaction successful"),
+         ok;
+    {'DOWN', MonRef, process, CompactPid, _Reason} ->
+         etap:bail("Failure compacting view group")
+    end,
+
+    ok = timer:sleep(2000),
+
+    etap:is(writer_try_again(Writer3), ok,
+        "Told writer 3 to try open his database again"),
+    etap:is(get_writer_status(Writer3), ok,
+        "Writer 3 was able to open his database"),
+
+    etap:is(is_process_alive(Writer1), true, "Writer 1 still alive"),
+    etap:is(is_process_alive(Writer2), true, "Writer 2 still alive"),
+    etap:is(is_process_alive(Writer3), true, "Writer 3 still alive"),
+
+    etap:is(stop_writer(Writer1), ok, "Stopped writer 1"),
+    etap:is(stop_writer(Writer2), ok, "Stopped writer 2"),
+    etap:is(stop_writer(Writer3), ok, "Stopped writer 3"),
+
+    delete_db(MainDb),
+    delete_db(DbWriter1),
+    delete_db(DbWriter2),
+    delete_db(DbWriter3).
+
+
+create_main_db() ->
+    {ok, Db} = create_db(main_db_name()),
+    DDoc = couch_doc:from_json_obj({[
+        {<<"_id">>, <<"_design/foo">>},
+        {<<"language">>, <<"javascript">>},
+        {<<"views">>, {[
+            {<<"foo">>, {[
+                {<<"map">>, <<"function(doc) { emit(doc._id, doc); }">>}
+            ]}},
+            {<<"foo2">>, {[
+                {<<"map">>, <<"function(doc) { emit(doc._id, doc); }">>}
+            ]}},
+            {<<"foo3">>, {[
+                {<<"map">>, <<"function(doc) { emit(doc._id, doc); }">>}
+            ]}},
+            {<<"foo4">>, {[
+                {<<"map">>, <<"function(doc) { emit(doc._id, doc); }">>}
+            ]}},
+            {<<"foo5">>, {[
+                {<<"map">>, <<"function(doc) { emit(doc._id, doc); }">>}
+            ]}}
+        ]}}
+    ]}),
+    {ok, _} = couch_db:update_doc(Db, DDoc, []),
+    ok = populate_main_db(Db, 1000, 20000),
+    update_view(Db#db.name, <<"_design/foo">>, <<"foo">>),
+    {ok, Db}.
+
+
+populate_main_db(Db, BatchSize, N) when N > 0 ->
+    Docs = lists:map(
+        fun(_) ->
+            couch_doc:from_json_obj({[
+                {<<"_id">>, couch_uuids:new()},
+                {<<"value">>, base64:encode(crypto:rand_bytes(1000))}
+            ]})
+        end,
+        lists:seq(1, BatchSize)),
+    {ok, _} = couch_db:update_docs(Db, Docs, []),
+    populate_main_db(Db, BatchSize, N - length(Docs));
+populate_main_db(_Db, _, _) ->
+    ok.
+
+
+update_view(DbName, DDocName, ViewName) ->
+    % Use a dedicated process -  we can't explicitly drop the #group ref counter
+    Pid = spawn(fun() ->
+        {ok, Db} = couch_db:open_int(DbName, []),
+        couch_view:get_map_view(Db, DDocName, ViewName, false),
+        ok = couch_db:close(Db)
+    end),
+    MonRef = erlang:monitor(process, Pid),
+    receive
+    {'DOWN', MonRef, process, Pid, normal} ->
+         etap:diag("View group updated"),
+         ok;
+    {'DOWN', MonRef, process, Pid, _Reason} ->
+         etap:bail("Failure updating view group")
+    end.
+
+
+create_db(DbName) ->
+    {ok, Db} = couch_db:create(
+        DbName,
+        [{user_ctx, #user_ctx{roles = [<<"_admin">>]}}, overwrite]),
+    {ok, Db}.
+
+
+delete_db(#db{name = DbName, main_pid = Pid}) ->
+    ok = couch_server:delete(
+        DbName, [{user_ctx, #user_ctx{roles = [<<"_admin">>]}}]),
+    MonRef = erlang:monitor(process, Pid),
+    receive
+    {'DOWN', MonRef, process, Pid, _Reason} ->
+        ok
+    after 30000 ->
+        etap:bail("Timeout deleting database")
+    end.
+
+
+spawn_writer(DbName) ->
+    Parent = self(),
+    spawn(fun() ->
+        process_flag(priority, high),
+        writer_loop(DbName, Parent)
+    end).
+
+
+get_writer_status(Writer) ->
+    Ref = make_ref(),
+    Writer ! {get_status, Ref},
+    receive
+    {db_open, Ref} ->
+        ok;
+    {db_open_error, Error, Ref} ->
+        Error
+    after 5000 ->
+        timeout
+    end.
+
+
+writer_try_again(Writer) ->
+    Ref = make_ref(),
+    Writer ! {try_again, Ref},
+    receive
+    {ok, Ref} ->
+        ok
+    after 5000 ->
+        timeout
+    end.
+
+
+stop_writer(Writer) ->
+    Ref = make_ref(),
+    Writer ! {stop, Ref},
+    receive
+    {ok, Ref} ->
+        ok
+    after 5000 ->
+        etap:bail("Timeout stopping writer process")
+    end.
+
+
+% Just keep the database open, no need to actually do something on it.
+writer_loop(DbName, Parent) ->
+    case couch_db:open_int(DbName, []) of
+    {ok, Db} ->
+        writer_loop_1(Db, Parent);
+    Error ->
+        writer_loop_2(DbName, Parent, Error)
+    end.
+
+writer_loop_1(Db, Parent) ->
+    receive
+    {get_status, Ref} ->
+        Parent ! {db_open, Ref},
+        writer_loop_1(Db, Parent);
+    {stop, Ref} ->
+        ok = couch_db:close(Db),
+        Parent ! {ok, Ref}
+    end.
+
+writer_loop_2(DbName, Parent, Error) ->
+    receive
+    {get_status, Ref} ->
+        Parent ! {db_open_error, Error, Ref},
+        writer_loop_2(DbName, Parent, Error);
+    {try_again, Ref} ->
+        Parent ! {ok, Ref},
+        writer_loop(DbName, Parent)
+    end.
diff --git a/test/etap/Makefile.am b/test/etap/Makefile.am
index c3a4ddab..313945d7 100644
--- a/test/etap/Makefile.am
+++ b/test/etap/Makefile.am
@@ -88,5 +88,6 @@ EXTRA_DIST = \
 	180-http-proxy.t \
   190-oauth.t \
 	200-view-group-no-db-leaks.t \
+	201-view-group-shutdown.t \
 	210-os-proc-pool.t