HDDS-5208. bump rocksdb version to 6.20.3

[JacksonYao287]

What changes were proposed in this pull request?

now , the rocksdb version used in ozone is 6.8.1, which is build on Apr 16 , 2020.
this PR bumps rockdsb to 6.20.3, which is build on May 5, 2021, and includes many new bug-fix

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-5208

How was this patch tested?

no need to add additional tests

[JacksonYao287]

@elek please take a look

[JacksonYao287]

@elek hi, in my local test , integration(ozone) test is passed, but it fail here. is it a flaky case?

[adoroszlai]

in my local test , integration(ozone) test is passed, but it fail here. is it a flaky case?

TestOzoneManagerRocksDBLogging has not been flaky recently, plus it looks related. I guess the problem could be that output may be changed in the new version.

(BTW, please trigger CI with empty commit instead of closing+reopening the PR. This makes it easier to see previous CI results.)

[JacksonYao287]

TestOzoneManagerRocksDBLogging has not been flaky recently, plus it looks related. I guess the problem could be that output may be changed in the new version.

ok, i will fix this

(BTW, please trigger CI with empty commit instead of closing+reopening the PR. This makes it easier to see previous CI results.)

thanks, will take care of this in the future

[JacksonYao287]

@adoroszlai i make the wait time longer ,and CI is passed. it seems we should wait longer for the log if we use the rocksdb of this version.

[elek]

thanks the patch @JacksonYao287

Retriggered the build to be sure that the (possible) intermittency is disappeared. If it passes, I am +1 to merge it.

[adoroszlai]

i make the wait time longer ,and CI is passed. it seems we should wait longer for the log if we use the rocksdb of this version.

Thanks for trying that.

The test expects the output db_impl.cc, which first appears when DUMPING STATS:

... [Thread-302] INFO  rocksdb.RocksDB (DBStoreBuilder.java:log(304)) - [/db_impl/db_impl.cc:932] ------- DUMPING STATS -------

Now with the new version there is a 15-sec delay. @avijayanhwx Would it make sense to check for rocksdb.RocksDB instead? I think it would appear in all lines coming from RocksDB:

ozone/hadoop-hdds/framework/src/main/java/org/apache/hadoop/hdds/utils/db/DBStoreBuilder.java

Lines 64 to 65 in 6b24c18

	public static final Logger ROCKS_DB_LOGGER =
	LoggerFactory.getLogger(RocksDB.class);

[avijayanhwx]

i make the wait time longer ,and CI is passed. it seems we should wait longer for the log if we use the rocksdb of this version.

Thanks for trying that.

The test expects the output db_impl.cc, which first appears when DUMPING STATS:

... [Thread-302] INFO  rocksdb.RocksDB (DBStoreBuilder.java:log(304)) - [/db_impl/db_impl.cc:932] ------- DUMPING STATS -------

Now with the new version there is a 15-sec delay. @avijayanhwx Would it make sense to check for rocksdb.RocksDB instead? I think it would appear in all lines coming from RocksDB:

ozone/hadoop-hdds/framework/src/main/java/org/apache/hadoop/hdds/utils/db/DBStoreBuilder.java

Lines 64 to 65 in 6b24c18

	public static final Logger ROCKS_DB_LOGGER =
	LoggerFactory.getLogger(RocksDB.class);

Yes, I am ok with that.

[JacksonYao287]

Retriggered the build to be sure that the (possible) intermittency is disappeared. If it passes, I am +1 to merge it.

thanks @elek for the retrigger。

[JacksonYao287]

Now with the new version there is a 15-sec delay. @avijayanhwx Would it make sense to check for rocksdb.RocksDB instead? I think it would appear in all lines coming from RocksDB:

@adoroszlai , in this test case , it seems that several rocksdb benchmarks are carried out and then the corresponding benchmark stats are dumped. so I think maybe in this version of rocksdb , more benchmarks are involved , and thus it takes longer time to complete

[adoroszlai]

Thanks @JacksonYao287 for the patch, @elek for the review. We can address the delay separately.