Saturday, November 18, 2017

How lsof Utility May Help MySQL DBAs

While working in Support, I noticed that probably at least once a week I have to use or mention lsof utility in some context. This week, for example, we had a customer trying to find out if his mysqld process running is linked with tcmalloc library. He started it different ways, using LD_PRELOAD directly and --malloc-lib option of mysqld_safe script etc, but wanted to verify that his attempts really worked as expected. My immediate comment in the internal chat was: "Just let them run lsof -p `pidof mysqld` | grep mall and check!" My MariaDB 10.2 instance uses jemalloc and this can be checked exactly the same way:
openxs@ao756:~/dbs/maria10.2$ ps aux | grep mysqld...
openxs    4619  0.0  0.0   4452   804 pts/2    S    17:02   0:00 /bin/sh bin/mysqld_safe --no-defaults --port=3308 --malloc-lib=/usr/lib/x86_64-linux-gnu/libjemalloc.so
openxs    4734  0.5  2.9 876368 115156 pts/2   Sl   17:02   0:00 /home/openxs/dbs/maria10.2/bin/mysqld --no-defaults --basedir=/home/openxs/dbs/maria10.2 --datadir=/home/openxs/dbs/maria10.2/data --plugin-dir=/home/openxs/dbs/maria10.2/lib/plugin --log-error=/home/openxs/dbs/maria10.2/data/ao756.err --pid-file=ao756.pid --port=3308
openxs    5391  0.0  0.0  14652   964 pts/2    S+   17:05   0:00 grep --color=auto mysqld
openxs@ao756:~/dbs/maria10.2$ lsof -p 4734 | grep mall
mysqld  4734 openxs  mem    REG              252,2    219776 12058822 /usr/lib/x86_64-linux-gnu/libjemalloc.so.1
openxs@ao756:~/dbs/maria10.2$
I think it's time to summarize most important use cases for lsof utility for MySQL DBAs. I am going to show different cases when it can be useful based on public MySQL bug reports. 

As one can read in the manual, lsof "lists on its standard output file information about files opened by processes". In one of the simplest possible calls presented above, we just pass PID of the process after -p option and get list of open files for this process. This includes shared libraries the process uses. By default the following format of the output is used:
openxs@ao756:~/dbs/maria10.2$ lsof -p 4734 | more
COMMAND  PID   USER   FD   TYPE             DEVICE  SIZE/OFF     NODE NAMEmysqld  4734 openxs  cwd    DIR              252,2      4096 29638597 /home/openxs/dbs/maria10.2/data
mysqld  4734 openxs  rtd    DIR              252,2      4096        2 /
mysqld  4734 openxs  txt    REG              252,2 147257671 29514843 /home/openxs/dbs/maria10.2/bin/mysqld
mysqld  4734 openxs  mem    REG              252,2     31792  1311130 /lib/x86_64-linux-gnu/librt-2.19.so
mysqld  4734 openxs  mem    REG              252,2 101270905 29241175 /home/openxs/dbs/maria10.2/lib/plugin/ha_rocksdb.so

...
mysqld  4734 openxs  DEL    REG               0,11             443265 /[aio]
mysqld  4734 openxs    0r   CHR                1,3       0t0     1050 /dev/null
mysqld  4734 openxs    1w   REG              252,2     52623  5255961 /home/openxs/dbs/maria10.2/data/.rocksdb/LOG
mysqld  4734 openxs    2w   REG              252,2    458880 29647192 /home/openxs/dbs/maria10.2/data/ao756.err
mysqld  4734 openxs    3r   DIR              252,2      4096  5255249 /home/openxs/dbs/maria10.2/data/.rocksdb
...
mysqld  4734 openxs  451u  IPv6             443558       0t0      TCP *:3308 (LISTEN)
mysqld  4734 openxs  452u  unix 0x0000000000000000       0t0   443559 /tmp/mysql.sock
...
mysqld  4734 openxs  470u   REG              252,2         0 29756970 /home/openxs/dbs/maria10.2/data/mysql/event.MYD
mysqld  4734 openxs  471u   REG              252,2         0 29647195 /home/openxs/dbs/maria10.2/data/multi-master.info
Most columns have obvious meaning, so let me concentrate on the few. FD should be numeric file descriptor, and it is for normal files. In this case it is also followed by a letter describing the mode under which the file is open (r for read, w for write and u for update). There may be one more letter describing a type of lock applied to the file. But we can see values without any single digit in the output above, so obviously some special values can be present there, like cwd for current working directory, rtd for root directory, txt for program text (code and data) or mem for memory mapped file, etc.

TYPE column is also interesting and may have plenty of values (as there are many types of files in Linux). REG means regular file, DIR is, obviously, a directory. Note also unix for a socket and IPv6 for the TCP port mysqld process listens to.

In SIZE/OFF column for normal files we usually see their size in bytes. Values for offset in file are usually prefixed with 0t if the value is decimal, or 0x if it's hex. NAME is obviously a fully specified file name (with symbolic links resolved). Some more details about the output format are discussed in the following examples.

Another usual way to use lsof is to pass a file name and get details about processes that have it opened, like this:
openxs@ao756:~/dbs/maria10.2$ lsof /tmp/mysql.sock
COMMAND  PID   USER   FD   TYPE             DEVICE SIZE/OFF   NODE NAME
mysqld  4734 openxs  452u  unix 0x0000000000000000      0t0 443559 /tmp/mysql.sock
openxs@ao756:~/dbs/maria10.2$ lsof /home/openxs/dbs/maria10.2
COMMAND     PID   USER   FD   TYPE DEVICE SIZE/OFF     NODE NAME
mysqld_sa  4619 openxs  cwd    DIR  252,2     4096 29235594 /home/openxs/dbs/maria10.2
lsof      14354 openxs  cwd    DIR  252,2     4096 29235594 /home/openxs/dbs/maria10.2
lsof      14355 openxs  cwd    DIR  252,2     4096 29235594 /home/openxs/dbs/maria10.2
bash      29244 openxs  cwd    DIR  252,2     4096 29235594 /home/openxs/dbs/maria10.2
In this case we see that /home/openxs/dbs/maria10.2 is used as a current working directory by 4 processes. Usually this kind of check is used when we can not unmount some directory, but it may be also useful in context of MySQL when you get error messages that some file is already used by other process. In the first example above I was checking what process could use /tmp/mysql.sock file.

Now, with the above details on basic usage in mind, let's check several recent enough MySQL bug reports that demonstrate typical and more advanced usage of lsof:
  • Bug #66237 - "Temporary files created by binary log cache are not purged after transaction commit". My former colleague and mentor from Percona, Miguel Angel Nieto (who recently joined a dark side of MongoDB employees) used lsof to show numerous files with names ML* created and left (until connection is closed) by mysqld process in /tmp directory (tmpdir to be precize) of a server with binary logging enabled, when transaction size was larger that binlog cache size. The bug is fixed in 5.6.17+ and 5.7.2+. It shows us a usual way of creating temporary files by MySQL server:
    # lsof -p 6112|grep ML
    mysqld  6112 root   38u   REG                7,0  106594304      18 /tmp/MLjw4ecJ (deleted)
    mysqld  6112 root   39u   REG                7,0  237314310      17 /tmp/MLwdWDGW (deleted)
    Notice (deleted) above. This is a result of immediate call to unlink() when temporary files are created. Check this in the source code, as well as my_delete() implementation.
  • Bug #82870 - "mysqld opens too many descriptors for slow query log". This bug (that is still "Verified") was opened by my former colleague Sveta Smirnova (now in Percona). Basically, mysqld opens too many descriptors for slow query log (and general query log) if it is turned ON and OFF while concurrent sessions are running. lsof allowed to see multiple descriptors created for the same file, until eventually open_files_limit is hit.
  • Bug #83434 - "Select statement with partition selection against MyISAM table opens all partitions". This bug (later declared a duplicate of older one and, eventually, a documented, even if unexpected, behavior by design) was opened by my colleague from MariaDB Geoff Montee. lsof utility helped to show that all partitions are actually opened by the mysqld process in this case.
  • Bug #74145 - "FLUSH LOGS improperly disables the logging if the log file cannot be accessed". This bug (still "Verified") was reported by Jean Weisbuch. Here we can see how lsof was used to find out if slow log is open after FLUSH. The logging has to be disabled, but MySQL continue to lie that it is enabled. I remember many cases when lsof also helped to find out where the error log (file with descriptor 2w) is really located/redirected to.
  • Bug #77752 - "bind-address wrongly prefers IPv4 over IPv6". This was not a bug (more like a configuration issue), but see how lsof -i is used by Daniël van Eeden to find out what process listens to a specific port, and does it listen to IPv4 or IPv6 address.
  • Bug #87589 - "Documentation incorrectly states that LOAD DATA LOCAL INFILE does not use tmpdir". In this "Verified" bug report Geoff Montee used lsof to show that temporary files are really created in tmpdir, not in /tmp (OS temporary directory). This is how you can find out when MySQL manual lies...
  • Bug #77519 - "Reported location of Innodb Merge Temp File is wrong". One more bug from Daniël van Eeden, this time "Verified". By calling lsof +L1 during an online alter table, he demonstrated that two temp files are created in tmpdir instead of in the datadir (as described by the manual), while events_waits_history_long table in performance_schema seems to claim it waited ion temporary file in the datadir. Note that in other his bug report, Bug #76225, fixed since 5.7.9 and 5.8.0, he had also shown ML* binlog cache files created that were not instrumented by performance_schema.
  • Bug #75706 - "alter table import tablespace creates a temporary table". This bug report by BJ Quinn is formally still "Verified", but according to my former colleague Przemyslaw Malkowski from Percona in recent 5.6.x and 5.7.x versions lsof does NOT show temporary table created. Time to re-verify this bug maybe, if the decision is made on how to implement this?
  • Bug #83717 - "Manual does not explain when ddl_log.log file is deleted and how large it can be". My own bug report where lsof was used to show that the ddl_log.log file remains open even after online ALTER completes. Manual is clear about this now.
To summarize, lsof may help MySQL DBA to find out:
  • what dynamic libraries are really used by the mysqld process
  • where the error log and other logs are really located
  • what other process may have some file, port or socket opened that is needed for current MySQL instance
  • why you may hit open_files_limit or use all free space in some filesystem unexpectedly
  • where all kinds of temporary files are created during specific operations
  • how MySQL really works with files, ports and sockets
It also allows to find MySQL bugs and clarify missing details in MySQL manual.

Saturday, November 11, 2017

Fun with Bugs #57 - On MySQL Bug Reports I am Subscribed to, Part I

I've decided to stop reviewing MySQL Release Notes in this series, but it does not mean that I am not interested in MySQL bugs any more. At the moment I am subscribed to 91 active MySQL bugs reported by other MySQL users, and in this blog post I am going to present 15 of them, the most recently reported ones. I'd really want to see them fixed or at least properly processed as soon as possible.

In some cases I am going to add my speculations on how the bug had better be handled, or maybe highlight some important details about it. It is not my job any more to process/"verify" any community bug reports for any kind of MySQL, but I did that for many years and I've spent more than 5 years "on the other side", being a member of Community, so in some cases I let myself to share some strong opinion on what may be done differently from the Oracle side.

As a side note, I started to subscribe to MySQL bugs mostly after I left Oracle, as before that I got email notification about each and every change in every MySQL bug report ever created...

Here is the list, starting from the most recent ones:
  • Bug #88422 - "MySQL 5.7 innodb purge thread get oldest readview could block other transaction". It is one of that bug reports without a test case from reporter. It is tempting to set it to "Verified" just "based on code review", as the code in 5.7 is quite obviously shows both holding the trx_sys->mutex and linear complexity of search depending on number of read views in the worst case (when most of them are closed):
    /**
    Get the oldest (active) view in the system.
    @return oldest view if found or NULL */

    ReadView*
    MVCC::get_oldest_view() const
    {
            ReadView*       view;

            ut_ad(mutex_own(&trx_sys->mutex));

            for (view = UT_LIST_GET_LAST(m_views);
                 view != NULL;
                 view = UT_LIST_GET_PREV(m_view_list, view)) {

                    if (!view->is_closed()) {
                            break;
                    }
            }

            return(view);
    }
    But probably current Oracle bugs verification rules do not let to just mark it as verified. After all, somebody will have to create a test case... So, my dear old friend Sinisa Milivojevic decided to try to force bug reporter to provide a test case instead of spending some time trying to create one himself. I am not going to blame him for that, why to try the easy way :) But I consider this his statement in the comment dated [10 Nov 16:21]:
    "... 5.7 methods holds no mutex what so ever..."
    a bit wrong, as we can see the mutex is acquired when get_oldest_view() method is called:
    void
    MVCC::clone_oldest_view(ReadView* view)
    {
            mutex_enter(&trx_sys->mutex);

            ReadView*       oldest_view = get_oldest_view();

            if (oldest_view == NULL) {
    ...
  • Bug #88381 - "Predicate cannot be pushed down "past" window function". Here bug reporter had provided enough hints for a test case. One can probably just check 'Handler%' status variables before and after query execution to come to the conclusion. Moreover, it seems Oracle developer,  Dag Wanvik, accepted this as a known limitation, but the bug still remains "Open" and nobody knows if it was copied to the internal bugs database, got prioritized and if any work on this is planned any time soon. We shell see. You may also want to monitor MDEV-10855.
  • Bug #88373 - "Renaming a column breaks replication from 5.7 to 8.0 because of impl. collation". This bug was quickly verified by Umesh Shastry. I expect a lot of "fun" for users upgrading to MySQL 8.0 when it becomes GA, especially in replication setups.
  • Bug #88328 - "Performance degradation with the slave_parallel_workers increase". There is no test case, just some general description and ideas about the root case when semi-sync replication is used. I expect this bug to stay "Open" for a long time, as it is a topic for a good research and blog posts like this one, that is, a work for real expert!
  • Bug #88223 - "Replication with no tmpdir space and InnoDB as tmp_storage_engine can break". Here we have clear and simple test case from Sveta Smirnova (no wonder, she also worked at bugs verification team in MySQL, Sun and Oracle). I hope Umesh will verify it soon. As a side note, it is explained (in the comments) elsewhere that InnoDB as internal_tmp_disk_storage_engine may not be the best possible option. We do not have this variable and do not plan to support InnoDB for internal temporary tables in MariaDB 10.2+.
  • Bug #88220 - "compressing and uncompressing InnoDB tables seems to be inconsistent". See also other, older bug reports mentioned there that are duplicates/closely related, but were not getting proper attention.
  • Bug #88150 - "'Undo log record is too big.' error occurring in very narrow range of str length". It was reported by my colleague Geoff Montee and is already fixed in recent versions of MariaDB (see MDEV-14051 for the details and some nice examples of gdb usage by a developer)!
  • Bug #88127 - "Index not used for 'order by' query with utf8mb4 character set". Here I am just curious when bugs like that would be caught up by Oracle QA before any public releases.
  • Bug #88071 - "An arresting Performance degradation when set sort_buffer_size=32M". here the test case is clear - just run sysbench oltp test at high concurrency with different values of sort_buffer_size. Still, Sinisa Milivojevic decided to explain when RAM limit may play a role instead of just showing how it works great (if it does) on any server with enough RAM... Let's see how this attempt to force bug reporter to work/explain more may end up...
  • Bug #87947 - "Optimizer chooses ref over range when access when range access is faster". Nice example of a case when optimizer trace may be really useful. Øystein Grøvlen kindly explained that "range access and ref access are not comparable costs". I wish we get better cost model for such cases in MySQL one day.
  • Bug #87837 - "MySQL 8 does not start after upgrade to 8.03". It is expected actually, and even somewhat documented in the release notes that MySQL 8.0.3 is not compatible to any older version. So, it is more like MySQL Installer (that I do not care much about) bug, but I still subscribed to it as yet another source of potential fun during further upgrade attempts.
  • Bug #87716 - "SELECT FOR UPDATE with BETWEEN AND gets row lock excessively". I think I already studied once why with IN() rows are locked differently by InnoDB comparing to BETWEEN that selects the same rows. But I'd like to know what's the Oracle's take on this, and I'd like to study this specific test case in details one day as well.
  • Bug #87670 - "Force index for group by is not always honored". Clear and simple test case, so no wonder it was immediately verified.
  • Bug #87621 - "Huge InnoDB slowdown when selecting strings without indexes ". I'd like to check with perf one day where the time is spent mostly during this test. For now I think this is a result of the way "long" data are stored on separate pages in InnoDB. What;'s interesting here is also a test case where R is used to generate data set.
  • Bug #87589 - "Documentation incorrectly states that LOAD DATA LOCAL INFILE does not use tmpdir". This was yet another report from my colleague Geoff Montee. lsof is your friend, maybe I have to talk about it one day at FOSDEM (call for papers is still open :) I like to find and follow bugs and missing details in MySQL manual, maybe because I would never be able to contribute to it as a writer directly...

So, this list shows my typical recent interests related to MySQL bugs - mostly InnoDB, optimizer, replication problems, fine manual and just some fun details like the way some Oracle engineers try to avoid working extra hard while processing bugs... I am also happy to know that in some cases MariaDB is able to deliver fixes faster.