Commit Graph

29 Commits

Author SHA1 Message Date
Yueh-Hsuan Chiang
fca5aa6fcc Initial script for the new regression test
Summary:
This diff includes an initial script running a set of benchmarks for
regression test.  The script does the following things:

  checkout the specified rocksdb commit (or origin/master as default)
  make clean && DEBUG_LEVEL=0 make db_bench
  setup test directories
  run set of benchmarks and store results

Currently, the script will run couple benchmarks, store all the benchmark
output, extract micros per op and percentile information for each benchmark
and store them in a single SUMMARY.csv file.  The SUMMARY.csv will make the
follow-up regression detection easier.

In addition, the current script only takes env arguments to set important
attributes of db_bench.  Will follow-up with a patch that allows db_bench
to construct options from an options file.

Test Plan:
NUM_KEYS=100 ./tools/regression_test.sh

  Sample SUMMARY.csv file:

                                     commit id,                      benchmark,  ms-per-op,        p50,        p75,        p99,      p99.9,     p99.99
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,                        fillseq,      15.28,      54.66,      77.14,    5000.00,   17900.00,   18483.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,                      overwrite,      13.54,      57.69,      86.39,    3000.00,   15600.00,   17013.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,                     readrandom,       1.04,       0.80,       1.67,     293.33,     395.00,     504.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,               readwhilewriting,       2.75,       1.01,       1.87,     200.00,     460.00,     485.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,                   deleterandom,       3.64,      48.12,      70.09,     200.00,     336.67,     347.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,                     seekrandom,      24.31,     391.87,     513.69,     872.73,     990.00,    1048.00
      7e23ddf575890510e7d2fc7a79b31a1bbf317917,         seekrandomwhilewriting,      14.02,     185.14,     294.15,     700.00,    1440.00,    1527.00

Reviewers: sdong, IslamAbdelRahman, kradhakrishnan, yiwu, andrewkr, gunnarku

Reviewed By: gunnarku

Subscribers: gunnarku, MarkCallaghan, andrewkr, dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D57597
2016-05-09 13:32:57 -07:00
Andrew Kryczka
4032145adc Configurable compression in db_bench
Summary:
Made compression type and dictionary size configurable via environment
variables.

Depends on D52287.

Test Plan:
check these options are passed to the db.

  $ COMPRESSION_MAX_DICT_BYTES=65536 COMPRESSION_TYPE=LZ4 NUM_KEYS=10000000 DB_DIR=./tmp/ WAL_DIR=./tmp/ ./tools/benchmark.sh filluniquerandom
  ...
  $ grep Options.compression tmp/LOG
  2016/04/22-19:11:30.397829 7f5f263a2980          Options.compression: LZ4
  ...
  2016/04/22-19:11:30.397837 7f5f263a2980         Options.compression_opts.max_dict_bytes: 65536

Reviewers: IslamAbdelRahman, sdong

Reviewed By: sdong

Subscribers: andrewkr, dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D57141
2016-04-27 17:39:18 -07:00
Andrew Kryczka
c3c389d542 Fix column label for L0 write sum
Summary:
This is taken from the "Write(GB)" column in compaction stats, so the
units should be GB, not MB.

Test Plan: none

Reviewers: sdong, yhchiang, IslamAbdelRahman

Reviewed By: IslamAbdelRahman

Subscribers: leveldb, andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D56889
2016-04-18 14:34:45 -07:00
Marton Trencseni
9b51987521 Adding pin_l0_filter_and_index_blocks_in_cache feature and related fixes.
Summary:
When a block based table file is opened, if prefetch_index_and_filter is true, it will prefetch the index and filter blocks, putting them into the block cache.
What this feature adds: when a L0 block based table file is opened, if pin_l0_filter_and_index_blocks_in_cache is true in the options (and prefetch_index_and_filter is true), then the filter and index blocks aren't released back to the block cache at the end of BlockBasedTableReader::Open(). Instead the table reader takes ownership of them, hence pinning them, ie. the LRU cache will never push them out. Meanwhile in the table reader, further accesses will not hit the block cache, thus avoiding lock contention.

Test Plan:
'export TEST_TMPDIR=/dev/shm/ && DISABLE_JEMALLOC=1 OPT=-g make all valgrind_check -j32' is OK.
I didn't run the Java tests, I don't have Java set up on my devserver.

Reviewers: sdong

Reviewed By: sdong

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D56133
2016-04-01 10:42:39 -07:00
sdong
b1fafcaca6 Revert "Adding pin_l0_filter_and_index_blocks_in_cache feature."
This reverts commit 522de4f59e.

It has bug of index block cleaning up.
2016-03-21 11:50:42 -07:00
Marton Trencseni
522de4f59e Adding pin_l0_filter_and_index_blocks_in_cache feature.
Summary:
When a block based table file is opened, if prefetch_index_and_filter is true, it will prefetch the index and filter blocks, putting them into the block cache.
What this feature adds: when a L0 block based table file is opened, if pin_l0_filter_and_index_blocks_in_cache is true in the options (and prefetch_index_and_filter is true), then the filter and index blocks aren't released back to the block cache at the end of BlockBasedTableReader::Open(). Instead the table reader takes ownership of them, hence pinning them, ie. the LRU cache will never push them out. Meanwhile in the table reader, further accesses will not hit the block cache, thus avoiding lock contention.
When the table reader is destroyed, it releases the pinned blocks (if there were any). This has to happen before the cache is destroyed, so I had to introduce a TableReader::Close(), to guarantee the order of destruction.

Test Plan:
Added two unit tests for this. Existing unit tests run fine (default is pin_l0_filter_and_index_blocks_in_cache=false).

DISABLE_JEMALLOC=1 OPT=-g make all valgrind_check -j32
  Mac: OK.
  Linux: with D55287 patched in it's OK.

Reviewers: sdong

Reviewed By: sdong

Subscribers: andrewkr, leveldb, dhruba

Differential Revision: https://reviews.facebook.net/D54801
2016-03-17 22:40:01 +00:00
Gunnar Kudrjavets
90aff0c444 Update --max_write_buffer_number for compaction benchmarks
Summary: For compactions benchmarks (both level and universal) we'll use `--max_write_buffer_number=4`. For all the other benchmarks which don't customize the value of `--max_background_flushes` we'll continue using `--max_write_buffer_number=8`.

Test Plan:
To validate basic correctness and command-line options:

```
cd ~/rocksdb
NKEYS=10000000 ./tools/run_flash_bench.sh
```

Reviewers: MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55497
2016-03-17 10:14:23 -07:00
Gunnar Kudrjavets
697fab820a Updates to RocksDB subcompaction benchmarking script
Summary: Set of updates to the subcompaction benchmark script which are based on our internal discussions. The intent behind the changes is to make sure that the scripts will correctly reflect how we're doing the actual benchmarking.

Test Plan: Tested by exercising the full set of compaction benchmarks and validating the execution and consistency of results.

Reviewers: MarkCallaghan, sdong, yhchiang

Reviewed By: yhchiang

Subscribers: andrewkr, dhruba

Differential Revision: https://reviews.facebook.net/D55461
2016-03-14 23:09:04 -07:00
Gunnar Kudrjavets
68189f7e1b Update benchmarks used to measure subcompaction performance
Summary: After closely working with Mark, Siying, and Yueh-Hsuan this set of changes reflects the updates needed to measure RocksDB subcompaction performance in a correct manner. The essence of the benchmark is executing `fillrandom` followed by `compact` with the correct set of options for various number of subcompactions specified.

Test Plan: Tested internally to verify correctness and reliability.

Reviewers: sdong, yhchiang, MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D55089
2016-03-04 12:32:11 -08:00
Gunnar Kudrjavets
337671b688 Add universal compaction benchmarks to run_flash_bench.sh
Summary:
Implement a benchmark for universal compaction based on the feature description (see below), in-person discussions, and reading source code:

https://github.com/facebook/rocksdb/wiki/RocksDB-Tuning-Guide
https://github.com/facebook/rocksdb/wiki/Universal-Compaction
https://github.com/facebook/rocksdb/wiki/RocksDB-Tuning-Guide#universal-compaction

Universal compaction benchmark is based on `overwrite` benchmark, adding compaction specific options to it, and executing it for different values of subcompaction to understand the impact of scaling out subcompactions for a particular scenario.

Test Plan:
  - Execute the benchmark on various machines for multiple iterations to verify the reliability.
  - Observe the output to make sure that compaction is taking place.
  - Observe the execution to make sure that arguments passed to `db_bench` are correct.

Reviewers: sdong, MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D54045
2016-02-10 15:30:47 -08:00
Gunnar Kudrjavets
8ed3438778 Add option to run fillseq with WAL enabled in addition to WAL disabled
Summary: This set of changes is part of the work to introduce benchmark for universal style compaction in RocksDB. It's conceptually separate from the compaction work, so sending it out as a separate diff to get it out of the way.

Test Plan:
  - Run `./tools/run_flash_bench.sh`.
  - Look at the contents of `report.txt` and `report2.txt` to make sure that data is reported and attributed correctly.
  - During `db_bench` execution time make sure that the correct flags are passed to `--disable_wal` depending on the benchmark being executed.

Reviewers: MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D53865
2016-02-05 13:20:56 -08:00
Mark Callaghan
4041903ecd Enhance db_bench write rate limit
Summary:
1) changes tools/{benchmark,run_flash_bench}.sh to optionally use the write rate limit
2) removes code for --writes_per_second and switches the 'background' write rate limit
to use --benchmark_write_rate_limit

Replaces https://reviews.facebook.net/D49113

Task ID: #9555881

Blame Rev:

Test Plan:
tools/run_flash_bench.sh

Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D52485
2016-01-04 12:01:27 -08:00
Mark Callaghan
4c81ac0c59 Fix benchmark report script
Summary:
db_bench output now displays Percentile many times with --statistics after
read IO latency histograms were added. So I only need the last one in the report output.

Task ID: #

Blame Rev:

Test Plan:
run run_flash_bench.sh

Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D45093
2015-08-22 12:18:00 -07:00
Mark Callaghan
41a0e2811d Improve defaults for benchmarks
Summary:
Changes include:
* don't sync-on-commit for single writer thread in readwhile... tests
* make default block size 8kb rather than 4kb to avoid too small blocks after compression
* use snappy instead of zlib to avoid stalls from compression latency
* disable statistics
* use bytes_per_sync=8M to reduce throughput loss on disk
* use open_files=-1 to reduce mutex contention

Task ID: #

Blame Rev:

Test Plan:
run benchmark

Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D44961
2015-08-20 18:59:10 -07:00
agiardullo
dc9d70de65 Optimistic Transactions
Summary: Optimistic transactions supporting begin/commit/rollback semantics.  Currently relies on checking the memtable to determine if there are any collisions at commit time.  Not yet implemented would be a way of enuring the memtable has some minimum amount of history so that we won't fail to commit when the memtable is empty.  You should probably start with transaction.h to get an overview of what is currently supported.

Test Plan: Added a new test, but still need to look into stress testing.

Reviewers: yhchiang, igor, rven, sdong

Reviewed By: sdong

Subscribers: adamretter, MarkCallaghan, leveldb, dhruba

Differential Revision: https://reviews.facebook.net/D33435
2015-05-29 14:36:35 -07:00
Mark Callaghan
88044340c1 Add Size-GB column to benchmark reports
Summary:
See https://gist.github.com/mdcallag/b867ee051d765760be0d for a sample

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D37971
2015-05-02 07:46:12 -07:00
Mark Callaghan
283a042969 Set --seed per test
Summary:
This is done to avoid having each thread use the same seed between runs
of db_bench. Without this we can inflate the OS filesystem cache hit rate on
reads for read heavy tests and generally see the same key sequences get generated
between teste runs.

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D37563
2015-04-23 09:18:25 -07:00
Mark Callaghan
78dbd087d1 Improve benchmark scripts
Summary:
This adds:
1) use of --level_compaction_dynamic_level_bytes=true
2) use of --bytes_per_sync=2M
The second is a big win for disks. The first helps in general.

This also adds a new test, fillseq with 32kb values to increase the peak
ingest and make it more likely that storage limits throughput.

Sample outpout from the first 3 tests - https://gist.github.com/mdcallag/e793bd3038e367b05d6f

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D37509
2015-04-22 13:23:08 -07:00
Mark Callaghan
9da8748016 Get benchmark.sh loads to run faster
Summary:
This changes loads to use vector memtable and disable the WAL. This also
increases the chance we will see IO bottlenecks during loads which is good to stress
test HW. But I also think it is a good way to load data quickly as this is a bulk
operation and the WAL isn't needed.

The two numbers below are the MB/sec rates for fillseq, bulkload using a skiplist
or vector memtable and the WAL enabled or disabled. There is a big benefit from
using the vector memtable and WAL disabled. Alas there is also a perf bug in
the use of std::sort for ordered input when the vector is flushed. Task is open
for that.
  112, 66 - skiplist with wal
  250, 116 - skiplist without wal
  110, 108 - vector with wal
  232, 370 - vector without wal

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D36957
2015-04-13 17:18:07 -07:00
Mark Callaghan
3be82bc894 Add p99.9 and p99.99 response time to benchmark report, add new summary report
Summary:
This adds p99.9 and p99.99 response times to the benchmark report and
adds a second report, report2.txt that has tests listed in test order rather
than the time in which they were run, so overwrite tests are listed for
all thread counts, then update etc.

Also changes fillseq to compress all levels to avoid write-amp from rewriting
uncompressed files when they reach the first level to compress.

Increase max_write_buffer_number to avoid stalls during fillseq and make
max_background_flushes agree with max_write_buffer_number.

See https://gist.github.com/mdcallag/297ff4316a25cb2988f7 for an example
of the new report (report2.txt)

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D36537
2015-04-06 10:42:12 -07:00
Mark Callaghan
1bd70fb54a Add --stats_interval_seconds to db_bench
Summary:
The --stats_interval_seconds determines interval for stats reporting
and overrides --stats_interval when set. I also changed tools/benchmark.sh
to report stats every 60 seconds so I can avoid trying to figure out a
good value for --stats_interval per test and per storage device.

Task ID: #6631621

Blame Rev:

Test Plan:
run tools/run_flash_bench, look at output

Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D36189
2015-03-30 12:58:32 -07:00
Mark Callaghan
99ec2412e5 Make the benchmark scripts configurable and add tests
Summary:
This makes run_flash_bench.sh configurable. Previously it was hardwired for 1B keys and tests
ran for 12 hours each. That kept me from using it. This makes it configuable, adds more tests,
makes the duration per-test configurable and refactors the test scripts.

Adds the seekrandomwhilemerging test to db_bench which is the same as seekrandomwhilewriting except
the writer thread does Merge rather than Put.

Forces the stall-time column in compaction IO stats to use a fixed format (H:M:S) which makes
it easier to scrape and parse. Also adds an option to AppendHumanMicros to force a fixed format.
Sometimes automation and humans want different format.

Calls thread->stats.AddBytes(bytes); in db_bench for more tests to get the MB/sec summary
stats in the output at test end.

Adds the average ingest rate to compaction IO stats. Output now looks like:
https://gist.github.com/mdcallag/2bd64d18be1b93adc494

More information on the benchmark output is at https://gist.github.com/mdcallag/db43a58bd5ac624f01e1

For benchmark.sh changes default RocksDB configuration to reduce stalls:
* min_level_to_compress from 2 to 3
* hard_rate_limit from 2 to 3
* max_grandparent_overlap_factor and max_bytes_for_level_multiplier from 10 to 8
* L0 file count triggers from 4,8,12 to 4,12,20 for (start,stall,stop)

Task ID: #6596829

Blame Rev:

Test Plan:
run tools/run_flash_bench.sh

Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D36075
2015-03-30 11:28:25 -07:00
Mark Callaghan
dfccc7b4e2 Add readwhilemerging benchmark
Summary:
This is like readwhilewriting but uses Merge rather than Put in the writer thread.
I am using it for in-progress benchmarks. I don't think the other benchmarks for Merge
cover this behavior. The purpose for this test is to measure read performance when
readers might have to merge results. This will also benefit from work-in-progress
to add skewed key generation.

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D35115
2015-03-18 13:50:52 -07:00
Mark Callaghan
58878f1c6a Switch to use_existing_db=1 for updaterandom and mergerandom
Summary:
Without this change about half of the updaterandom reads and merge puts will be for keys that don't exist.
I think it is better for these tests to start with a full database and use fillseq to fill it.

Task ID: #

Blame Rev:

Test Plan:
Revert Plan:

Database Impact:

Memcache Impact:

Other Notes:

EImportant:

- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -

Reviewers: igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D35043
2015-03-14 08:36:57 -07:00
Leonidas Galanis
e126e0da5b Single threaded tests -> sync=0 Multi threaded tests -> sync=1 by default unless DB_BENCH_NO_SYNC is defined
Summary:
Single threaded tests -> sync=0 Multi threaded tests -> sync=1 by default unless DB_BENCH_NO_SYNC is defined.

Also added updaterandom and mergerandom with putOperator. I am waiting for some results from udb on this.

Test Plan:
DB_BENCH_NO_SYNC=1 WAL_DIR=/tmp OUTPUT_DIR=/tmp/b DB_DIR=/tmp ./tools/benchmark.sh debug,bulkload,fillseq,overwrite,filluniquerandom,readrandom,readwhilewriting,updaterandom,mergerandom

WAL_DIR=/tmp OUTPUT_DIR=/tmp/b DB_DIR=/tmp ./tools/benchmark.sh debug,bulkload,fillseq,overwrite,filluniquerandom,readrandom,readwhilewriting,updaterandom,mergerandom

Verify sync settings

Reviewers: sdong, MarkCallaghan, igor, rven

Reviewed By: igor, rven

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D34185
2015-03-06 14:12:53 -08:00
Leonidas Galanis
9d5bd411be benchmark.sh won't run through all tests properly if one specifies wal_dir to be different than db directory.
Summary:
A command line like this to run all the tests:
source benchmark.config.sh && nohup ./benchmark.sh 'bulkload,fillseq,overwrite,filluniquerandom,readrandom,readwhilewriting'
where
benchmark.config.sh is:
export DB_DIR=/data/mysql/rocksdata
export WAL_DIR=/txlogs/rockswal
export OUTPUT_DIR=/root/rocks_benchmarking/output

Will fail for the tests that need a new DB .

Also 1) set disable_data_sync=0 and 2) add debug mode to run through all the tests more quickly

Test Plan: run ./benchmark.sh 'debug,bulkload,fillseq,overwrite,filluniquerandom,readrandom,readwhilewriting' and verify that there are no complaints about WAL dir not being empty.

Reviewers: sdong, yhchiang, rven, igor

Reviewed By: igor

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D30909
2015-01-05 15:36:47 -08:00
Lei Jin
e93f044d99 add range scan test to benchmark script
Summary: as title

Test Plan: ran it

Reviewers: yhchiang, igor, sdong, MarkCallaghan

Reviewed By: MarkCallaghan

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D25563
2014-12-10 13:04:58 -08:00
Lei Jin
f18b4a4847 minor update to benchmark script
Summary: Try to match some parameters from Dhruba's benchmarks on github

Test Plan: ran it

Reviewers: sdong, yhchiang, igor

Reviewed By: igor

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D24687
2014-10-10 09:55:40 -07:00
Lei Jin
add22e3515 standardize scripts to run RocksDB benchmarks
Summary:
Hope these scripts will allow people to run/repro benchmark easily
I think it is time to re-run flash benchmarks and report results
Please comment if any other benchmark runs are needed

Test Plan: ran it

Reviewers: yhchiang, igor, sdong

Reviewed By: igor

Subscribers: dhruba, MarkCallaghan, leveldb

Differential Revision: https://reviews.facebook.net/D23139
2014-09-12 16:25:35 -07:00