tag:blogger.com,1999:blog-16959946.post7196271050740995899..comments2023-12-09T16:44:47.897+01:00Comments on The Data Charmer: MariaDB 10 is a Sandbox killjoy?Giuseppe Maxiahttp://www.blogger.com/profile/15801583338057324813noreply@blogger.comBlogger8125tag:blogger.com,1999:blog-16959946.post-82296926248400288442014-11-12T06:58:42.227+01:002014-11-12T06:58:42.227+01:00perhaps trying sandbox from latest source could he...perhaps trying sandbox from latest source could helpjosephhttp://javadevnotes.comnoreply@blogger.comtag:blogger.com,1999:blog-16959946.post-31070036429853815972014-06-05T17:28:36.651+02:002014-06-05T17:28:36.651+02:00Server should be safe to use as much as it would b...Server should be safe to use as much as it would be safe if you switched off native AIO manually in InnoDB options. It is still a real configuration <b>error</b> and should be noticed by the DBA who can then decide what to do about it: fix, get rid of the message by unsetting the option manually, or just ignore. <br /><br />There is no particular reason to limit server to only having "warnings" or "fatal errors". <br /><br />In MySQL it was a fatal error. In MariaDB, it became a non-fatal error. I don't see why a DBA should have a problem with this. <br /><br />If the DBA wants to treat every error as a fatal error, they are free to do so just as you do in sandbox. Otherwise they have an option to filter more-serious-but-still-not-life-threatening problems out of the whole universe of warnings that the server can produce, and act accordingly. <br /><br />For 5.6 vs 10.0 comparison, we'll investigate why MySQL doesn't produce the error anymore, whether there was a real fix, or the message is just suppressed, or it's something else. If it's a real fix that escaped the merge, it should be incorporated. If it was just a cosmetic change to have a nice-looking error, then it's better to keep a clear error message in MariaDB. elensthttps://www.blogger.com/profile/16293598411807790232noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-37438092871004444092014-06-05T11:01:48.667+02:002014-06-05T11:01:48.667+02:00@elenst:
It's just not a fatal error.
Does it...@elenst:<br /><i>It's just not a fatal error.</i><br /><br />Does it mean that the server is safe to use, even though it says there is an error during bootstrap? <br /><br />DBAs evaluating the software must be thrilled.<br /><br />If it is not serious, it should say "WARNING." <br />It is bad enough that the server is mixing useless messages and errors in STDERR. If the errors being listed have different meaning, this is terrible for usability.<br /><br />Giuseppe Maxiahttps://www.blogger.com/profile/15801583338057324813noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-36097465566520806342014-06-05T10:24:05.438+02:002014-06-05T10:24:05.438+02:00It is an error, and the server is right saying so....It is an error, and the server is right saying so. It's just not a fatal error. elensthttps://www.blogger.com/profile/16293598411807790232noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-40629711076589369862014-06-05T01:57:33.192+02:002014-06-05T01:57:33.192+02:00@Elenst,
the only setting that MySQL::Sandbox appl...@Elenst,<br />the only setting that MySQL::Sandbox applies to the database when running multiple sandboxes are server-id and binary logs activation.<br /><br />The host is Ubuntu 12.04 with 16GB or RAM.<br /><br />You are right about MySQL Sandbox stopping if it detects an error in the output. This is due to MySQL 5.6 and MariaDB 10 <a href="http://datacharmer.blogspot.ca/2013/12/old-and-new-mysql-verbosity.html" rel="nofollow">inconsistent behavior during bootstrap</a>. Up to version 5.5, bootstrap was sending its output to STDOUT, and only errors to STDERR. <br />That way, it was easy to detect if there was ane error. With the current behavior, all output goes to STDERR, which forces me to parse for the word "error". So, if the database server says "error" in STDERR the application can legitimately assume that there was an error. If there isn't an error but the server says "error," then it's a different kind of bug.Giuseppe Maxiahttps://www.blogger.com/profile/15801583338057324813noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-83959536317473810382014-06-05T00:04:22.727+02:002014-06-05T00:04:22.727+02:00I took a very quick look at the sandbox code; I co...I took a very quick look at the sandbox code; I could be wrong, but it looks like sandbox aborts bootstrap if it finds in output anything that resembles an error. It could account for the strange server shutdown.<br /><br />But then, there should be nothing new for sandbox here. MySQL used to just abort in this case and it was a documented limitation (http://bugs.mysql.com/bug.php?id=61575), then about 2 years ago in MariaDB 5.5 the fatal error was replaced with the error message, so MariaDB 5.5 would throw the same warning/error and continue, and the outcome for sandbox should be the same as now. <br /><br />It is worth checking though why the error still happens with XtraDB, while with InnoDB it does not. elensthttps://www.blogger.com/profile/16293598411807790232noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-77379967639438224612014-06-04T22:30:26.602+02:002014-06-04T22:30:26.602+02:00Have you, by any chance, tried MariaDB with InnoDB...Have you, by any chance, tried MariaDB with InnoDB plugin and Percona 5.6? <br /><br />The AIO warning (in MariaDB server) should be relatively harmless, it means what it says, native AIO gets disabled, and InnoDB continues initializing. It was a workaround for MySQL problem when InnoDB would abort in the same scenario. Apparently, it was finally fixed in 5.6.<br /><br />The log looks weird, it seems that the server just shuts down after InnoDB initialized. I started 25 MariaDB servers with default options on my Wheezy VM (manually, not via sandbox) and still didn't get this effect. If you share more details about the machine (OS, memory) and MariaDB server configuration, I will try to reproduce it on a similar VM. elensthttps://www.blogger.com/profile/16293598411807790232noreply@blogger.comtag:blogger.com,1999:blog-16959946.post-89019395535165066652014-06-04T09:04:29.385+02:002014-06-04T09:04:29.385+02:00Works perfectly on Fedora 20 Heisenburg. From that...Works perfectly on Fedora 20 Heisenburg. From that we could guestimate that CentOS is fine also.<br /><br />Run same command <i>make_multiple_sandbox --how_many_nodes=10 10.0.11</i> after unpacking Maria to ~/opt/mysql/10.0.11Rautiainenhttp://indeksed.netnoreply@blogger.com