|
--user USER --password --host ADDRESS --port PORT --socket SOCKET --no-mycnf --db (--Database) --help |
These options mimic most standard applications.
--no-mycnf makes mysqlsla not read ~/.my.cnf which it does by default otherwise.
--Database can be abbreviated -D (like the mysql cli). |
|
--general LOG --slow LOG --raw LOG |
One of these options is required. Each option refers to its respective type
of log. To specify multiple logs, comma separate the list of log files. For example,
to have mysqlsla read two general logs: --general file1.log,file2.log. mysqlsla
can read any amount and combination of logs so something like this is valid:
--general file1.log --slow file2.log,file3.log --raw file4.log. If mysqlsla can't
read a log it prints a warning and continues reading the other logs. At present,
all valid queries from all logs are combined; the analyses cannot be ran on
a per-log basis. General and slow logs are made by MySQL and mysqlsla should have
no problem reading any variation of them, but raw logs are made by other means,
therefore a specific format is required of them: SQL statements must be semicolon
terminated and new-line separated. A single SQL statement can be split across
lines, but its last line must be semicolon terminated and the next SQL statement
must start on a new line. Blank lines and lines that start with # or -- are skipped.
|
| --beautify |
As of mysqlsla v1.3, all SQL statements are "flattened," which means they
are set to all lowercase. This helps improve query abstraction because otherwise
"SELECT * FROM foo;" and "select * from foo;" appear to be different
statements in Perl (because Perl hash keys are case-sensitive). Unfortunately,
flattened statements aren't pretty read, so this options causes mysqlsla to
capitalize most important SQL keywords like SELECT, FROM, WHERE, ORDER BY, etc.
|
| --examples |
For all analyses except --print-queries and --time-each-query, mysqlsla prints the
abstracted form of queries. This options causes mysqlsla to print random,
not-abstracted, examples of the queries instead.
|
| --print-queries (--pq) |
Print all valid queries from all logs. This is useful if you want to see what
mysqlsla actually read from the logs. This option is not affected by --top and
is not a valid option to --correlate.
|
| --frequency (--fq) |
Print the most frequently occurring queries in descending order. The frequency
of each query is listed by count and percentage of all queries. This option is
affected by --top and is a valid option to --correlate.
|
| --explain (--ex) |
Print the queries with the largest potential result set of rows in descending
order. A query's potential result set of rows is the product of all rows for
all tables given by EXPLAIN, therefore this is only an estimate based on table
statistics. ANALYZE updates a table's statistics. A query with a large result
set often indicates improper or absent indexes. This option is affected by
--rows-read and --top and is a valid option to --correlate.
|
| --time-each-query (--te) |
Time how long it takes each query to execute and print in descending order.
No status indicator is given so if a query takes a really long time mysqlsla
might appear to be frozen but it's really just waiting for the query to finish.
This option is affected by --avg and --top and is a valid option to --correlate.
|
| --time-all-queries (--ta) |
Time how long it takes all queries to execute. This analysis has only
one result: the average total execution time of all queries. This option
is affected by --avg and --percent: the current time run (or percentage
complete of all time runs) is printed while the queries are being timed.
The resulting average is an average of all the time runs.
This option is not affected by --top and is not a valid option to --correlate.
|
|
--slow-time (--st, --at) --slow-lock (--sl, --al) --slow-rows-exam (--sre, --ar) --slow-rows-sent (--srs) |
These analyses work only with slow logs and they directly mimic the sort (-s) options of
mysqldumpslow (except --slow-rows-sent). If used with the --mysqldumpslow (--mds) option,
the output is nearly identical to mysqldumpslow. Missing here is the
mysqldumpslow sort option 'c' (count); that option is analogous to mysqlsla analysis
--frequency (--fq). For all these analyses, the values are averages for each unique
query, printed in descending order. These analyses are affected by --top and are valid
options to --correlate (as st, sl, sre, and srs, not at, al, ar).
|
|
--slow-total-time (--stt, --t) --slow-total-lock (--stl, --l) --slow-total-rows-exam (--stre, --r) --slow-total-rows-sent (--strs) |
These analyses work only with slow logs and they directly mimic the sort (-s) options of
mysqldumpslow (except --slow-total-rows-sent). If used with the --mysqldumpslow (--mds)
option, the output is nearly identical to mysqldumpslow. For all
these analyses, the values are totals for each unique query, printed in descending
order. These analyses are affected by --top and are valid options to --correlate (as
stt, stl, stre, strs, not t, l, r).
|
| --order A |
The default order the analyses are ran and printed is: pq, fq, ex, te, ta.
A new order can be specified with this option. The option's value A refers
to the analyses' aliases, comma separated. For example, to reverse the
default order: --order ta,te,ex,fq,pq. If an analyses is not specified in
the new order it is appended respective to its place in the default order,
however this unspecified analysis is not ran unless explicitly specified
by its own option. For example the following runs only analyses fq and ex:
--order fq,ex. Whereas the following run fq, ex, then pq, te because
pq and te are unspecified in the order, but explicitly specified, and in
the default order pq comes before te: --order fq,ex --pq --te.
(Internally mysqlsla makes the order fq,ex,pq,te,ta.)
When an analyses is given in --order its own option is
implied so the following is redundant: --order fq,ex --fq --ex.
Specifying an analysis more than once in the order works but is not
supported. For example, the following causes 3 frequency analyses
followed by 2 explain analyses: --order fq,fq,fq,ex,ex.
|
| --correlate A |
By default mysqlsla runs each analysis on the queries independently.
Therefore each analyses usually lists the queries differently. For example
query X can be the most frequent, but query Y produces the most rows in
its result set, and yet query Z is the slowest. To correlate analyses
means to use the results from one analysis in subsequent analyses.
This preserves the order of queries across subsequent analyses.
For example, if the analyses are correlated to the frequency analysis,
with the default order of analyses, subsequent analyses ex and te will
use the results of fq. So if analysis fq results in queries X, Y, Z, analysis
ex will list results for queries X, Y, Z although Y has the largest result
set, and analysis te will also list results for queries X, Y, Z although
Z has the longest execution time. Correlating results makes it easy to
determine things like "how long do the top 3 most frequent queries take
to execute" and "how frequently do the top 3 slowest queries occur."
These questions can be answered without correlation if you can manage
to find the same query in each analyses since the analyses will list
the queries in different orders. With correlation, the Nth query
in the correlated analyses is the Nth query in subsequent analyses.
— This option's value A refers to fq, ex, or te. If --correlate
is specified but the analysis not explicitly specified by its own
option or in --order, the analysis is implied and set first in the order overriding
--order. If the analysis is specified in --order, mysqlsla follows
--order. For example, the following correlates on ex and the order is
ex, fq: --correlate ex --order fq. But the follow correlates on
ex and the order is fq, ex: --correlate ex --order fq, ex. The order
is important because only subsequent analyses are correlated. If
the correlating analysis is last nothing correlates to it. For example,
the following run analyses ex and te independently then fq:
--correlate fq --order ex,te,fq. In general, an analysis cannot
correlate to another analysis which hasn't ran yet.
|
| --hide |
This option takes a comma-separated list of analysis
aliases (fq, te, st, etc.) and causes their output to
be suppressed but still executes them (if the analysis
is actually executed by its own option or --order or
--correlate). This can be used to show only the final
analysis. For example, to show only the execution time
of the top 3 most frequent queries: --correlate fq --order fq,te --hide fq --top 3.
|
| --percent |
For --time-each-query and --time-all-queries mysqlsla by default makes 1 time run.
If more time runs are specified with --avg, these time runs can be counted as a
percentage complete with this option. This only affects --time-all-queries (--time-each-query
doesn't indicate which time run its currently on). The percentage complete is
listed as %25, 50%, 75%. This option is implicitly invoked if 20 or more time
runs are specified with --avg.
|
| --rows-read (--rr) |
By default --explain calculates rows produced; this option makes it calculate
rows read. Read the article
JOIN Rows Produced vs. Rows Read
for more information about this distinction.
|
| --avg N (--n) |
For --time-each-query and --time-all-queries mysqlsla by default makes 1 time run.
This option causes mysqlsla to make N time runs and average the results. If N is
20 or more, --percent is automatically invoked.
|
| --flush-qc |
This option causes mysqlsla to FLUSH QUERY CACHE; before starting the analyses.
|
| --top N |
By default mysqlsla lists all queries for all analysis (except --time-all-queries).
This option limits the output to the top N queries.
|
| --filter S |
By default mysqlsla allows the following SQL
statements: DELETE, DO, INSERT, REPLACE, SELECT,
TRUNCATE, UPDATE, USE, CALL, SHOW, ROLLBACK, COMMIT.
It discards SET, START.
This is new as of v1.2; previous versions only allowed
SELECT and USE. Any SQL statement not recognized
(whether filtered or not) is discarded. This is to
prevent reading junk statements. The default filter can
be changed with this option. The parameter S is a
comma-separated list of above SQL statements or *
preceded by + to allow the statement or - to discard
the statement. For example, to allow only INSERT and
UPDATE: --filter -*,+INSERT,+UPDATE. To discard everything:
--filter -*. To discard only CALL:
--filter -CALL,+SET. To discard DELETE and DO and allow
SET: --filter -DELETE,-DO,+SET.
|
| --safe |
This overrides any --filter with: -*,+SELECT,+USE. This
option can be used to analyze an unknown log safely
because it will discard any statements that may modify
any database or table.
|
| --grep P |
This option causes mysqlsla to keep only statements
that match the Perl regular expression pattern P (case
insensitive). It is applied after statement filtering.
Since P is put directly into a pattern match (m/P/io),
you may need to escape some special characters in the
pattern like parenthesis.
|
| --mysqldumpslow (--mds) |
The --slow-* analyses use a mysqlsla format by default.
This option changes the format of the --slow-* analyses
and --frequency to look like mysqldumpslow.
|