From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Mon, 12 Nov 2018 16:03:34 +0300 From: Vladimir Davydov Subject: Re: [tarantool-patches] [PATCH v1 1/1] box: restore region after box.snapshot() Message-ID: <20181112130334.izjt4i7kvgvm3zjc@esperanza> References: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: To: imeevma@tarantool.org Cc: tarantool-patches@freelists.org, v.shpilevoy@tarantool.org List-ID: On Sat, Nov 03, 2018 at 05:11:42PM +0300, imeevma@tarantool.org wrote: > Before this patch region wasn't restored after box.snapshot() so > some tests fail due to "memory leak". After this patch region will > be restored using region_truncate(). It could be done by cleaning > memory in case there wasn't active transaction but it seems like > region_truncate() is more universal. > > Closes #3732 > --- > Issue: https://github.com/tarantool/tarantool/issues/3732 > Branch: https://github.com/tarantool/tarantool/tree/imeevma/gh-3732-restore_region_after_box_snapshot > > src/box/box.cc | 2 ++ > test/engine/snapshot.result | 35 +++++++++++++++++++++++++++++++++++ > test/engine/snapshot.test.lua | 17 +++++++++++++++++ > 3 files changed, 54 insertions(+) > > diff --git a/src/box/box.cc b/src/box/box.cc > index c4b4fb4..0d27aa2 100644 > --- a/src/box/box.cc > +++ b/src/box/box.cc > @@ -2157,6 +2157,7 @@ box_checkpoint() > if (! is_box_configured) > return 0; > int rc = 0; > + size_t used = region_used(&fiber()->gc); > if (box_checkpoint_is_in_progress) { > diag_set(ClientError, ER_CHECKPOINT_IN_PROGRESS); > return -1; > @@ -2181,6 +2182,7 @@ end: > > latch_unlock(&schema_lock); > box_checkpoint_is_in_progress = false; > + region_truncate(&fiber()->gc, used); > return rc; > } IMHO this looks confusing, because we typically do region_truncate only in functions/modules that actually allocate something on the region. Here one has to go much deeper to understand what's going on (I assume it's memtx_engine_begin_checkpoint) so chances are high somebody will try to delete this useless code sooner or later. We could probably add a comment here to clear things up, but if memtx_engine_begin_checkpoint is changed one day, we will definitely forget to remove the comment that will become obsolete. I'd prefer if you did region_truncate where objects are actually allocated or not use the region allocator at all - it doesn't seem to be a hot path after all.