[PATCH] [replication] [recovery] recover missing data
Konstantin Belyavskiy
k.belyavskiy at tarantool.org
Mon Mar 26 18:22:23 MSK 2018
Version for 1.6 without test (to add or not is under discussion)
Recover missing local data from replica.
In case of sudden power-loss, if data was not written to WAL but
already sent to remote replica, local can't recover properly and
we have different datasets.
Fix it by using remote replica's data and LSN comparison.
Based on @GeorgyKirichenko proposal
Closes #3210
---
branch: gh-3210-recover-missing-local-data-master-master-16
src/box/relay.cc | 12 +++-
src/box/relay.h | 5 ++
test/replication/master1.lua | 30 +++++++++
test/replication/master2.lua | 1 +
test/replication/recover_missing.result | 103 ++++++++++++++++++++++++++++++
test/replication/recover_missing.test.lua | 38 +++++++++++
6 files changed, 186 insertions(+), 3 deletions(-)
create mode 100644 test/replication/master1.lua
create mode 120000 test/replication/master2.lua
create mode 100644 test/replication/recover_missing.result
create mode 100644 test/replication/recover_missing.test.lua
diff --git a/src/box/relay.cc b/src/box/relay.cc
index 8e53ca022..092ea0627 100644
--- a/src/box/relay.cc
+++ b/src/box/relay.cc
@@ -250,6 +250,7 @@ relay_subscribe(int fd, struct xrow_header *packet,
* and identify ourselves with our own server id.
*/
struct xrow_header row;
+ relay.masters_lsn_at_subscribe = vclock_get(master_vclock, r->server_id);
xrow_encode_vclock(&row, master_vclock);
/*
* Identify the message with the server id of this
@@ -286,10 +287,15 @@ relay_send_row(struct recovery *r, void *param, struct xrow_header *packet)
* (JOIN request). In this case, send every row.
* Otherwise, we're feeding a WAL, thus responding to
* SUBSCRIBE request. In that case, only send a row if
- * it is not from the same server (i.e. don't send
- * replica's own rows back).
+ * it is not from the same server (i.e. don't send replica's
+ * own rows back). There is a special case if this row is
+ * missing on the other side (i.e. in case of sudden power-loss,
+ * data was not written to WAL, so remote master can't recover
+ * it). In this case packet's LSN is lower or equal then local
+ * master's LSN at the moment it has issued 'SUBSCRIBE' request.
*/
- if (packet->server_id == 0 || packet->server_id != r->server_id) {
+ if (packet->server_id == 0 || packet->server_id != r->server_id ||
+ packet->lsn <= relay->masters_lsn_at_subscribe) {
relay_send(relay, packet);
ERROR_INJECT(ERRINJ_RELAY,
{
diff --git a/src/box/relay.h b/src/box/relay.h
index c13267134..9cf7cbfac 100644
--- a/src/box/relay.h
+++ b/src/box/relay.h
@@ -45,6 +45,11 @@ struct relay {
uint64_t sync;
struct recovery *r;
ev_tstamp wal_dir_rescan_delay;
+ /**
+ * Local master's LSN at the moment of subscribe, used to check
+ * dataset on the other side and send missing data rows if any.
+ */
+ int64_t masters_lsn_at_subscribe;
};
/**
diff --git a/test/replication/master1.lua b/test/replication/master1.lua
new file mode 100644
index 000000000..758874ee7
--- /dev/null
+++ b/test/replication/master1.lua
@@ -0,0 +1,30 @@
+#!/usr/bin/env tarantool
+
+-- get instance name from filename (master1.lua => master1)
+local INSTANCE_ID = string.match(arg[0], "%d")
+local USER = 'cluster'
+local PASSWORD = 'somepassword'
+local SOCKET_DIR = require('fio').cwd()
+local function instance_uri(instance_id)
+ --return 'localhost:'..(3310 + instance_id)
+ return SOCKET_DIR..'/master'..instance_id..'.sock';
+end
+
+-- start console first
+require('console').listen(os.getenv('ADMIN'))
+
+box.cfg({
+ listen = instance_uri(INSTANCE_ID);
+ replication_source = {
+ USER..':'..PASSWORD..'@'..instance_uri(1);
+ USER..':'..PASSWORD..'@'..instance_uri(2);
+ };
+})
+
+box.once("bootstrap", function()
+ local test_run = require('test_run').new()
+ box.schema.user.create(USER, { password = PASSWORD })
+ box.schema.user.grant(USER, 'replication')
+ box.schema.space.create('test', {engine = test_run:get_cfg('engine')})
+ box.space.test:create_index('primary')
+end)
diff --git a/test/replication/master2.lua b/test/replication/master2.lua
new file mode 120000
index 000000000..f6ea42dd7
--- /dev/null
+++ b/test/replication/master2.lua
@@ -0,0 +1 @@
+master1.lua
\ No newline at end of file
diff --git a/test/replication/recover_missing.result b/test/replication/recover_missing.result
new file mode 100644
index 000000000..f49d2dc49
--- /dev/null
+++ b/test/replication/recover_missing.result
@@ -0,0 +1,103 @@
+env = require('test_run')
+---
+...
+test_run = env.new()
+---
+...
+SERVERS = { 'master1', 'master2' }
+---
+...
+-- Start servers
+test_run:create_cluster(SERVERS)
+---
+...
+-- Check connection status
+-- first on master 1
+test_run:cmd("switch master1")
+---
+- true
+...
+fiber = require('fiber')
+---
+...
+while box.info.replication.status ~= 'follow' do fiber.sleep(0.001) end
+---
+...
+box.info.replication.status
+---
+- follow
+...
+-- and then on master 2
+test_run:cmd("switch master2")
+---
+- true
+...
+fiber = require('fiber')
+---
+...
+while box.info.replication.status ~= 'follow' do fiber.sleep(0.001) end
+---
+...
+box.info.replication.status
+---
+- follow
+...
+test_run:cmd("switch master1")
+---
+- true
+...
+box.snapshot()
+---
+- ok
+...
+box.space.test:insert({1})
+---
+- [1]
+...
+box.space.test:count()
+---
+- 1
+...
+test_run:cmd("switch master2")
+---
+- true
+...
+box.space.test:count()
+---
+- 1
+...
+test_run:cmd("stop server master1")
+---
+- true
+...
+fio = require('fio')
+---
+...
+fio.unlink(fio.pathjoin(fio.abspath("."), string.format('master1/%020d.xlog', 0)))
+---
+- true
+...
+test_run:cmd("start server master1")
+---
+- true
+...
+test_run:cmd("switch master1")
+---
+- true
+...
+box.space.test:count()
+---
+- 1
+...
+test_run:cmd("switch default")
+---
+- true
+...
+test_run:cmd("stop server master1")
+---
+- true
+...
+test_run:cmd("stop server master2")
+---
+- true
+...
diff --git a/test/replication/recover_missing.test.lua b/test/replication/recover_missing.test.lua
new file mode 100644
index 000000000..964056786
--- /dev/null
+++ b/test/replication/recover_missing.test.lua
@@ -0,0 +1,38 @@
+env = require('test_run')
+test_run = env.new()
+
+SERVERS = { 'master1', 'master2' }
+-- Start servers
+test_run:create_cluster(SERVERS)
+
+-- Check connection status
+-- first on master 1
+test_run:cmd("switch master1")
+fiber = require('fiber')
+while box.info.replication.status ~= 'follow' do fiber.sleep(0.001) end
+box.info.replication.status
+-- and then on master 2
+test_run:cmd("switch master2")
+fiber = require('fiber')
+while box.info.replication.status ~= 'follow' do fiber.sleep(0.001) end
+box.info.replication.status
+
+test_run:cmd("switch master1")
+box.snapshot()
+box.space.test:insert({1})
+box.space.test:count()
+
+test_run:cmd("switch master2")
+box.space.test:count()
+test_run:cmd("stop server master1")
+fio = require('fio')
+fio.unlink(fio.pathjoin(fio.abspath("."), string.format('master1/%020d.xlog', 0)))
+test_run:cmd("start server master1")
+
+test_run:cmd("switch master1")
+box.space.test:count()
+
+test_run:cmd("switch default")
+test_run:cmd("stop server master1")
+test_run:cmd("stop server master2")
+
--
2.14.3 (Apple Git-98)
More information about the Tarantool-patches
mailing list