Tarantool development patches archive
 help / color / mirror / Atom feed
From: Serge Petrenko <sergepetrenko@tarantool.org>
To: Vladislav Shpilevoy <v.shpilevoy@tarantool.org>,
	tarantool-patches@dev.tarantool.org, gorcunov@gmail.com
Subject: Re: [Tarantool-patches] [PATCH 10/12] raft: move box_update_ro_summary to update trigger
Date: Tue, 17 Nov 2020 15:42:27 +0300	[thread overview]
Message-ID: <85c52e01-c876-8d8e-1543-bf272efb1d79@tarantool.org> (raw)
In-Reply-To: <637105222f13010a9ed0488ee698b2ba2684b46e.1605570907.git.v.shpilevoy@tarantool.org>


17.11.2020 03:02, Vladislav Shpilevoy пишет:
> box_update_ro_summary() was called from the basic Raft library,
> making it depend on box. But it was called every time when Raft
> state was changed and broadcasted. It means the same effect can be
> achieved by updating RO summary from Raft state update trigger.
>
> The patch does it, and now Raft code does not depend on box.h.
>
> Part of #5303
> ---
>   src/box/raft.c    | 2 ++
>   src/box/raftlib.c | 8 --------
>   2 files changed, 2 insertions(+), 8 deletions(-)
>
> diff --git a/src/box/raft.c b/src/box/raft.c
> index f3652bbcb..db1a3f423 100644
> --- a/src/box/raft.c
> +++ b/src/box/raft.c
> @@ -77,6 +77,8 @@ box_raft_on_update_f(struct trigger *trigger, void *event)
>   	(void)trigger;
>   	struct raft *raft = (struct raft *)event;
>   	assert(raft == box_raft());
> +	/* State or enablence could be changed, affecting read-only state. */
> +	box_update_ro_summary();
>   	if (raft->state != RAFT_STATE_LEADER)
>   		return 0;
>   	/*


Raft uses synchronous WAL write, corect?

So there's a yield in raft_worker_handle_io(). Now there's a period of 
time when
an instance is a follower, but it isn't read-only.

When you reconfigure a leader to become voter, everything's fine, since no
writing to disk is involved.

However, if an existing leader receives a message with term greater, 
than its own,
it'll first persist this term, and thus yield, and proceed to broadcast 
and switching
to ro later.

So now it's possible that a follower is writeable for some period of time.

Maybe run on_update triggers before the wal write? Even better, run the 
triggers
on the actual state transition. After each  `raft->state = ...`.

On the bright side, your patch seems to fix
https://github.com/tarantool/tarantool/issues/5440

> diff --git a/src/box/raftlib.c b/src/box/raftlib.c
> index 512dbd51f..2e09d5405 100644
> --- a/src/box/raftlib.c
> +++ b/src/box/raftlib.c
> @@ -33,7 +33,6 @@
>   #include "error.h"
>   #include "fiber.h"
>   #include "small/region.h"
> -#include "box.h"
>   #include "tt_static.h"
>   
>   /**
> @@ -603,8 +602,6 @@ raft_sm_become_leader(struct raft *raft)
>   	raft->state = RAFT_STATE_LEADER;
>   	raft->leader = raft->self;
>   	ev_timer_stop(loop(), &raft->timer);
> -	/* Make read-write (if other subsystems allow that. */
> -	box_update_ro_summary();
>   	/* State is visible and it is changed - broadcast. */
>   	raft_schedule_broadcast(raft);
>   }
> @@ -655,7 +652,6 @@ raft_sm_schedule_new_term(struct raft *raft, uint64_t new_term)
>   	raft->volatile_vote = 0;
>   	raft->leader = 0;
>   	raft->state = RAFT_STATE_FOLLOWER;
> -	box_update_ro_summary();
>   	raft_sm_pause_and_dump(raft);
>   	/*
>   	 * State is visible and it is changed - broadcast. Term is also visible,
> @@ -686,7 +682,6 @@ raft_sm_schedule_new_election(struct raft *raft)
>   	/* Everyone is a follower until its vote for self is persisted. */
>   	raft_sm_schedule_new_term(raft, raft->term + 1);
>   	raft_sm_schedule_new_vote(raft, raft->self);
> -	box_update_ro_summary();
>   }
>   
>   static void
> @@ -771,7 +766,6 @@ raft_sm_start(struct raft *raft)
>   		 */
>   		raft_sm_wait_leader_found(raft);
>   	}
> -	box_update_ro_summary();
>   	/*
>   	 * Nothing changed. But when raft was stopped, its state wasn't sent to
>   	 * replicas. At least this was happening at the moment of this being
> @@ -793,7 +787,6 @@ raft_sm_stop(struct raft *raft)
>   		raft->leader = 0;
>   	raft->state = RAFT_STATE_FOLLOWER;
>   	ev_timer_stop(loop(), &raft->timer);
> -	box_update_ro_summary();
>   	/* State is visible and changed - broadcast. */
>   	raft_schedule_broadcast(raft);
>   }
> @@ -879,7 +872,6 @@ raft_cfg_is_candidate(struct raft *raft, bool is_candidate)
>   			raft_schedule_broadcast(raft);
>   		}
>   	}
> -	box_update_ro_summary();
>   }
>   
>   void

-- 
Serge Petrenko

  reply	other threads:[~2020-11-17 12:42 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-11-17  0:02 [Tarantool-patches] [PATCH 00/12] Raft module, part 2 - relocation to src/lib/raft Vladislav Shpilevoy
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 01/12] raft: move sources to raftlib.h/.c Vladislav Shpilevoy
2020-11-17  8:14   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 10/12] raft: move box_update_ro_summary to update trigger Vladislav Shpilevoy
2020-11-17 12:42   ` Serge Petrenko [this message]
2020-11-17 15:17     ` Serge Petrenko
2020-11-18 23:21     ` Vladislav Shpilevoy
2020-11-19 10:08       ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 11/12] raft: introduce RaftError Vladislav Shpilevoy
2020-11-17 15:13   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 12/12] raft: move algorithm code to src/lib/raft Vladislav Shpilevoy
2020-11-17 15:13   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 02/12] raft: move box_raft_* to src/box/raft.h and .c Vladislav Shpilevoy
2020-11-17  8:14   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 03/12] raft: stop using replication_disconnect_timeout() Vladislav Shpilevoy
2020-11-17  8:15   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 04/12] raft: stop using replication_synchro_quorum Vladislav Shpilevoy
2020-11-17  8:17   ` Serge Petrenko
2020-11-19 23:42     ` Vladislav Shpilevoy
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 05/12] raft: stop using instance_id Vladislav Shpilevoy
2020-11-17  8:59   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 06/12] raft: make raft_request.vclock constant Vladislav Shpilevoy
2020-11-17  9:17   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 07/12] raft: stop using replicaset.vclock Vladislav Shpilevoy
2020-11-17  9:23   ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 08/12] raft: introduce vtab for disk and network Vladislav Shpilevoy
2020-11-17  9:35   ` Serge Petrenko
2020-11-19 23:43     ` Vladislav Shpilevoy
2020-11-17 10:00   ` Serge Petrenko
2020-11-19 23:43     ` Vladislav Shpilevoy
2020-11-20  7:56       ` Serge Petrenko
2020-11-20 19:40         ` Vladislav Shpilevoy
2020-11-23  8:09           ` Serge Petrenko
2020-11-17  0:02 ` [Tarantool-patches] [PATCH 09/12] raft: introduce raft_msg, drop xrow dependency Vladislav Shpilevoy
2020-11-17 10:22   ` Serge Petrenko
2020-11-19 23:43     ` Vladislav Shpilevoy
2020-11-20  8:03       ` Serge Petrenko

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=85c52e01-c876-8d8e-1543-bf272efb1d79@tarantool.org \
    --to=sergepetrenko@tarantool.org \
    --cc=gorcunov@gmail.com \
    --cc=tarantool-patches@dev.tarantool.org \
    --cc=v.shpilevoy@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH 10/12] raft: move box_update_ro_summary to update trigger' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox