From: Vladislav Shpilevoy <v.shpilevoy@tarantool.org> To: tarantool-patches@freelists.org, Evgeniy Zaikin <evgen4rave@gmail.com> Subject: [tarantool-patches] Re: [PATCH V3] Prohibit router/storage configuration from different fibers Date: Thu, 20 Sep 2018 21:59:21 +0200 [thread overview] Message-ID: <823876a0-0bfb-e119-36d6-525b08372297@tarantool.org> (raw) In-Reply-To: <CAHwoe2Wii98Nep8C7bSZtKFcXYTt7pnrxvqz_qKgi8DRXwkNDQ@mail.gmail.com> Hi! Sorry for the last response - I was on vacation last 2 weeks. See 8 comments below. On 11/09/2018 12:32, Evgeniy Zaikin wrote: > From 5ce83d5db595455a14a58aedf9e6411f5adf1d7c Mon Sep 17 00:00:00 2001 > From: rave4life <evgen4rave@gmail.com> > Date: Fri, 31 Aug 2018 16:41:11 +0300 > Subject: [PATCH V3] Prohibit router/storage configuration from different fibers > > Fixes: #140 > --- > Ticket:https://github.com/tarantool/vshard/issues/140 > Branch:https://github.com/tarantool/vshard/tree/rave4life/gh-140-prohibit-multiple-cfg > > diff --git a/test/router/router.result b/test/router/router.result > index 000307d..ada8309 100644 > --- a/test/router/router.result > +++ b/test/router/router.result > @@ -1140,6 +1140,45 @@ test_run:drop_cluster(REPLICASET_2) > while test_run:grep_log('router_1', 'disconnected from ') == nil do > fiber.sleep(0.1) end > --- > ... > +-- gh-140: prohibit vshard.router/storage.cfg from multiple fibers. 1. Sorry, still out of 80 symbols. It is a strict rule. Please, inspect your patch carefully for such code style things before sending to a review. > +cfg = {sharding = require('localcfg').sharding} > +--- > +... > +vshard = require('vshard') > +--- > +... > +fiber = require('fiber') 2. Please, do not use 'default' server for testing storage or router. Router/router.test.lua starts multiple Tarantool instances: storage_1_a - storage_2_b and router_1 (see the very beginning of the test file) and switches between them to test router or storage things. 'Default' server shall not be used for this. Please, use test_run:switch and router_1 server for a test. In such a case you will not have require('vshard') and require('fiber') since they are already loaded on router_1 instance. There is a reason why 'default' can not be used - once a router is started on an instance, it can not be stopped without restart (its internal fibers, I mean). For the storage things are worse - it creates new spaces, users, functions and we have no way to cleanup those things. So once 'default' is used for testing, it will affect next tests. For this problem we have an issue: https://github.com/tarantool/vshard/issues/121 You can try it next after this patch is finished if you want. > +--- > +... > +vshard.router.cfg(cfg) > +--- > +... > +vshard.router.internal.errinj.ERRINJ_MULTIPLE_CFG = false 3. It is false by default already (as I see in router/init.lua), so what is a point of this line? I removed it and the test passed ok. I guess it is a typo. Please, fix. > +--- > +... > +c = fiber.channel(2) > +--- > +... > diff --git a/test/storage/storage.result b/test/storage/storage.result > index e3e9e7f..0228445 100644 > --- a/test/storage/storage.result > +++ b/test/storage/storage.result > @@ -688,6 +688,48 @@ rs:callro('echo', {'some_data'}) > - null > - null > ... > +-- gh-140: prohibit vshard.router/storage.cfg from multiple fibers. > +cfg = {sharding = {}} > +--- > +... > +cfg.sharding[box.info.cluster.uuid] = {replicas = {}} > +--- > +... > +cfg.sharding[box.info.cluster.uuid].replicas[box.info.uuid] = {uri = > 'storage:storage@127.0.0.1:3309', name = 'test_storage', master = > true} > +--- > +... > +vshard = require('vshard') > +--- > +... > +fiber = require('fiber') 4. For the same reason I explained above, you do not need 'require' here. This code is executed on 'storage_1_a' where both fiber and vshard are loaded already (see line storage.test.lua:137 - this is the latest switch before your test). > +--- > +... > +vshard.storage.internal.errinj.ERRINJ_MULTIPLE_CFG = false 5. Same as 3. > +--- > +... > +c = fiber.channel(2) > +--- > +... > diff --git a/vshard/router/init.lua b/vshard/router/init.lua > index 7573801..446b1a0 100644 > --- a/vshard/router/init.lua > +++ b/vshard/router/init.lua > @@ -28,6 +28,7 @@ if not M then > ---------------- Common module attributes ---------------- > errinj = { > ERRINJ_CFG = false, > + ERRINJ_MULTIPLE_CFG = false, > ERRINJ_FAILOVER_CHANGE_CFG = false, > ERRINJ_RELOAD = false, > ERRINJ_LONG_DISCOVERY = false, > @@ -53,6 +54,10 @@ local ROUTER_TEMPLATE = { > name = nil, > -- The last passed configuration. > current_cfg = nil, > + -- Fiber channel as the locker to prevent multiple config. > + cfg_lock = nil, > + -- Configuration counter to control and verify locker. > + cfg_counter = 0, 6. Please, remove cfg_counter. Its value > 0 is unreachable during normal work of router/storage. If you need a counter for testing lock, you can use erring.ERRINJ_MULTIPLE_CFG as a counter. It is not necessary to make each errinj as a strict boolean. See my proposal below in the comment 8. > -- Time to outdate old objects on reload. > connection_outdate_delay = nil, > -- Bucket map cache. > @@ -594,6 +599,29 @@ local function router_cfg(router, cfg, is_reload) > end > end > > +-------------------------------------------------------------------------------- > +-- Configuration wrapper > +-------------------------------------------------------------------------------- 7. Function comment should look like this: -- -- My function description wrapped by 66 -- symbols. -- @param name1 Parameter 1 description. -- @param name2 Parameter 2 description. -- -- [if returns anything] -- @retval Returned value description. -- @retval Next possible return value description. -- > + > +local function cfg_wrapper(router, ...) > + local enable_lock = not M.errinj.ERRINJ_MULTIPLE_CFG > + if enable_lock then > + router.cfg_lock:put(true) > + end > + if router.cfg_counter > 0 then > + error('Error: cfg_counter > 0') > + end > + router.cfg_counter = router.cfg_counter + 1 > + local status, err = pcall(router_cfg, router, ...) > + router.cfg_counter = router.cfg_counter - 1 > + if enable_lock then > + router.cfg_lock:get() > + end > + if not status then > + error(err) > + end > +end 8. Too complex. Lets use errinj.ERRINJ_MULTIPLE_CFG as an integer and set it to nil by default M = { ... errinj = { ... ERRINJ_MULTIPLE_CFG = nil } ... } ... function cfg_wrapper() if ERRINJ_MULTIPLE_CFG then ERRINJ_MULTIPLE_CFG = ERRINJ_MULTIPLE_CFG + 1 if ERRINJ_MULTIPLE_CFG > 1 then error('Multiple cfgs at the same time') end status, err = pcall(router_cfg, router, ...) ERRINJ_MULTIPLE_CFG = ERRINJ_MULTIPLE_CFG - 1 else lock:put(true) status, err = pcall(router_cfg, router, ...) lock:get() end if not status then error(err) end return end ERRINJ_MULTIPLE_CFG is nil by default and to active it you can do this in a test: ERRINJ_MULTIPLE_CFG = 0 This will trigger counter in my pseudo code above since '0' in Lua is true.
prev parent reply other threads:[~2018-09-20 19:59 UTC|newest] Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top 2018-09-11 10:32 [tarantool-patches] " Evgeniy Zaikin 2018-09-20 19:59 ` Vladislav Shpilevoy [this message]
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=823876a0-0bfb-e119-36d6-525b08372297@tarantool.org \ --to=v.shpilevoy@tarantool.org \ --cc=evgen4rave@gmail.com \ --cc=tarantool-patches@freelists.org \ --subject='[tarantool-patches] Re: [PATCH V3] Prohibit router/storage configuration from different fibers' \ /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