Lack of free pages in the buffer pool is not diagnosed with innodb_empty_free_list_algorithm=backoff

[Laurynas Biveinis]

If innodb_empty_free_list_algorithm=backoff (which is the default), then the following code is potentially skipped in buf_LRU_get_free_block:

1) Error log diagnostics and InnoDB monitor trigger:

                fprintf(stderr,
                        " InnoDB: Warning: difficult to find free blocks in\n"
                        "InnoDB: the buffer pool (%lu search iterations)!\n"
                        "InnoDB: %lu failed attempts to flush a page!"
                        " Consider\n"
                        "InnoDB: increasing the buffer pool size.\n"
                        "InnoDB: It is also possible that"
                        " in your Unix version\n"
                        "InnoDB: fsync is very slow, or"
                        " completely frozen inside\n"
                        "InnoDB: the OS kernel. Then upgrading to"
                        " a newer version\n"
                        "InnoDB: of your operating system may help."
                        " Look at the\n"
                        "InnoDB: number of fsyncs in diagnostic info below.\n"
                        "InnoDB: Pending flushes (fsync) log: %lu;"
                        " buffer pool: %lu\n"
                        "InnoDB: %lu OS file reads, %lu OS file writes,"
                        " %lu OS fsyncs\n"
                        "InnoDB: Starting InnoDB Monitor to print further\n"
                        "InnoDB: diagnostics to the standard output.\n",
                        (ulong) n_iterations,
                        (ulong) flush_failures,
                        (ulong) fil_n_pending_log_flushes,
                        (ulong) fil_n_pending_tablespace_flushes,
                        (ulong) os_n_file_reads, (ulong) os_n_file_writes,
                        (ulong) os_n_fsyncs);

                mon_value_was = srv_print_innodb_monitor;
                started_monitor = TRUE;
                srv_print_innodb_monitor = TRUE;
                os_event_set(lock_sys->timeout_event);

2) Innodb_buffer_pool_wait_free status variable and associated InnoDB monitor bump:

        srv_stats.buf_pool_wait_free.add(n_iterations, 1);

[Laurynas Biveinis (laurynas-biveinis)]

https://github.com/percona/percona-server/pull/1291
https://github.com/percona/percona-server/pull/1292

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1775