| Bug #49781 | Insufficient restart types for API nodes listed in ConfigInfo.cpp | ||
|---|---|---|---|
| Submitted: | 17 Dec 2009 19:30 | Modified: | 4 Jan 2010 10:39 |
| Reporter: | Matthew Montgomery | Email Updates: | |
| Status: | Not a Bug | Impact on me: | |
| Category: | MySQL Cluster: Cluster (NDB) storage engine | Severity: | S4 (Feature request) |
| Version: | mysql-5.1-telco-7.0 | OS: | Any |
| Assigned to: | Magnus Blåudd | CPU Architecture: | Any |
| Tags: | 7.0.10 | ||
[17 Dec 2009 19:35]
MySQL Verification Team
RSA type is needed when adding (or moving?) MGM nodes in a live cluster also.
[18 Dec 2009 21:23]
Magnus Blåudd
The restart type is defined in the comment for enum ConfigInfo::Flags in ConfigInfo.hpp and the default is as you see "node restart of one node at a time for the setting to take effect" and that means ALL(ndb_mgmd, ndbd and api) nodes.
We have identified a couple of parameters that if changed need initial or even initial system restart and have marked those appropriately. We currently have _no_ parameters that does not need any restart at all.
Sorry, but I must close this as not a bug.
"
enum Flags {
CI_ONLINE_UPDATEABLE = 1, // Parameter can be updated online
CI_CHECK_WRITABLE = 2, // Path given by parameter should be writable
/*
Flags telling how the system must be restarted for a changed
parameter to take effect
Default is none of these flags set, which means node restart
of one node at a time for the setting to take effect
CS_RESTART_INITIAL
Each data node need to be restarted one at a time with --initial
CS_RESTART_SYSTEM
The whole system need to be stopped and then started up again
CS_RESTART_SYSTEM + CS_RESTART_INITIAL
The whole system need to be stopped and then restarted with --initial
thus destroying any data in the cluster
These flags can not be combined with CI_ONLINE_UPDATABLE flag which
indicates that the parameter can be changed online without
restarting anything
*/
CI_RESTART_SYSTEM = 4, // System restart is necessary to apply setting
CI_RESTART_INITIAL = 8 // Initial restart is necessary to apply setting
};"
[4 Jan 2010 10:39]
Jon Stephens
Anything changed in config.ini requires at a minimum a rolling restart of the cluster to take effect. This is the default. These facts are documented. Any change that happens or appears to work without performing at least a rolling restart does just that -- happens or appears to work -- and is not guaranteed behaviour. Therefore, I'm closing this as !BUG as suggested by Magnus.
[8 Jan 2010 9:58]
Jon Stephens
Furthermore: It's not appropriate to have info not relating to Cluster parameters in the output of ndb_config. Info about MySQL server options should be obtained from `mysqld --help --verbose` (or by some other means from mysqld.cc). MCM should determine whether an attribute originates with config.ini or my.cnf. In the latter case, it should restart only the mysqld to which that my.cnf belongs.
Description: ./storage/ndb/src/mgmsrv/ConfigInfo.cpp Shows restart types of "default" or simply a Node restart for Hostname and ExecuteOnComputer changes on [API] nodes. However just restarting the lone API node is not sufficient. For either of these changes rolling restart of the ndb_mgmd and cluster ndb(mt)d nodes and that API node is required. However there is no restart type defined to indicate that a rolling system restart is required in addition to the single node restart. The current list of restart types N, IN, S and IS do not sufficiently cover to all possible restart types. How to repeat: 1776 /*************************************************************************** 1777 * API 1778 ***************************************************************************/ 1804 { 1805 CFG_NODE_HOST, 1806 "HostName", 1807 API_TOKEN, 1808 "Name of computer for this node", 1809 ConfigInfo::CI_USED, 1810 false, 1811 ConfigInfo::CI_STRING, 1812 "", 1813 0, 0 }, 1850 { 1851 KEY_INTERNAL, 1852 "ExecuteOnComputer", 1853 API_TOKEN, 1854 "String referencing an earlier defined COMPUTER", 1855 ConfigInfo::CI_USED, 1856 false, 1857 ConfigInfo::CI_STRING, 1858 UNDEFINED, 1859 0, 0 }, Suggested fix: Restart types should be defined and documented as follows: N : Node (Just this MGM, NDBD or API node) IN : Initial Node (Just this MGM or NDBD node, wiping all data) S : System (All MGM then all NDBD nodes at the same time) IS : Initial System (All MGM then all NDBD nodes at the same time, wiping all data) RSN: Rolling System+Node (All MGM and NDBD nodes one after another and just this API node) RSA: Rolling System+All API Nodes (All MGM and NDBD nodes one after another and all API nodes) RSN type is needed when adding or moving API nodes in a live cluster. RSA type is needed when adding (or moving?) NDBD nodes in a live cluster. This additional clarity of this restart type can help management applications (MCM) make more appropriate restart type choices. Also See: Bug #47366