THIS IS A TEST INSTANCE. ALL YOUR CHANGES WILL BE LOST!!!!
1. stable builds/intermittent test failures
There are no intermittent failures specific to the HDDS-2823 branch as of now. During the development , it was ensured all the CI checks are clean prior to every commit merge .The plan is to run repeated CI checks on the merge commit to master.
2. documentation
All the docs are available in the doc section DOCS.
TODO: add it to the docs (ozone.apache.org/docs/)
3. design, attached the docs
All the design docs are also available in the doc section DOCS.
TODO: add it to the docs (ozone.apache.org/docs/)
4. s3 compatibility
There are no incompatibilities with respect to s3. S3 is not changed at all (except that the OmUtils rename affected it)
5. support of containers / kubernetes:
Tested in Kubernetes and everything worked well. Example files are uploaded in the patch to:
6. coverage/code quality:
Sonar master branch.
Sonar SCM-HA branch
SCM-HA has 20 more issues but 2.3 better code coverage.
7. build time
Recent master builder: https://github.com/apache/ozone/actions/runs/638668950
Recent SCM-HA build: https://github.com/apache/ozone/actions/runs/638165382
SCM-HA branch didn't introduce any significant slowness (2-3 minutes plus to the existing integrations test and acceptance tests which are already close to 1h runtime).
There is no significant difference between local build time. In a linux with the below configuration,
hadoop@9 ~/glengeng$ lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 16 On-line CPU(s) list: 0-15 Thread(s) per core: 1 Core(s) per socket: 16 Socket(s): 1 NUMA node(s): 1 Vendor ID: GenuineIntel CPU family: 6 Model: 94 Model name: Intel(R) Xeon(R) Gold 61xx CPU Stepping: 3 CPU MHz: 2494.138 BogoMIPS: 4988.27 Hypervisor vendor: KVM Virtualization type: full L1d cache: 32K L1i cache: 32K L2 cache: 4096K NUMA node0 CPU(s): 0-15 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl eagerfpu pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch fsgsbase bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap xsaveopt xsavec xgetbv1 arat hadoop@9 ~/glengeng$ free -g total used free shared buff/cache available Mem: 125 4 98 0 22 120 Swap: 0 0 0 #data for master INFO BUILD SUCCESS INFO --------------------------------------------- INFO Total time: 08:10 min INFO Finished at: 2021-03-10T16:54:42+08:00 INFO --------------------------------------------- #data for HDDS-2823 INFO BUILD SUCCESS INFO -------------------------------------------- INFO Total time: 07:09 min INFO Finished at: 2021-03-10T17:06:42+08:00
8. possible incompatible changes/used feature flag:
For using the SCM HA feature, "ozone.scm.ratis.enable" config needs to be set to be true in ozone-site.xml.
9. third party dependencies/licence changes:
Checking the content of the two branches (find -type f | sort > ... + diff) the only jar differences are due to a latest version bump:
< ./share/ozone/lib/jackson-annotations-2.10.3.jar < ./share/ozone/lib/jackson-core-2.10.3.jar < ./share/ozone/lib/jackson-databind-2.10.3.jar < ./share/ozone/lib/jackson-dataformat-cbor-2.10.3.jar < ./share/ozone/lib/jackson-dataformat-xml-2.10.3.jar < ./share/ozone/lib/jackson-datatype-jsr310-2.10.3.jar < ./share/ozone/lib/jackson-module-jaxb-annotations-2.10.3.jar --- > ./share/ozone/lib/jackson-annotations-2.12.1.jar > ./share/ozone/lib/jackson-core-2.12.1.jar > ./share/ozone/lib/jackson-databind-2.12.1.jar > ./share/ozone/lib/jackson-dataformat-cbor-2.12.1.jar > ./share/ozone/lib/jackson-dataformat-xml-2.12.1.jar > ./share/ozone/lib/jackson-datatype-jsr310-2.12.1.jar > ./share/ozone/lib/jackson-module-jaxb-annotations-2.12.1.jar
No new dependencies are added.
10. performance
Performance between master and SCM-HA branch (without turning on Ratis) is shared here
We use the default configuration for master and 2823.
The write throughput seems to be constrained by hardware, e.g. DC network, which we haven’t dug further.
According to the slight differences between 2823 and master, the performance of the SCM HA bypass Ratis is close to that of pure in-mem SCM.