China 2020 PGConf Asia 2020 Developer Unconference
China 2020 PGConf Asia 2020 Developer Unconference Meeting Notes
The developer Unconference for “CHINA 2020 And PGConf.Asia 2020” took place on Nov 17th at 9am Beijing time, it was an online Unconference using the zoom online meeting tool. The meeting was very well attended, PostgreSQL experts and enthusiast joined the conference around the globe despite the inconvenient time for some participants.
Here are they highlights of the meeting :
- The topic selection and voting process was managed by Bruce Momjain and Grant, the topics were submitted for both English and Maradian language.
- The following international PG experts and community members joined the developer Unconference :
- Bruce Momjain (EnterpriseDB) - Masahiko Swada (EnterpriseDB) - Amit Langote (EnterpriseDB) - Hemmi-san (NTTOSS) - Fujita-san (NTTOSS) - Andrey Lepikhov (PostgrePro) - Ahsan Hadi (HighGO CA) - Muhammad Usama (HighGO CA) - Hamid Akhtar (HighGO CA) - Cary (HighGO CA) - Asif Rehman (HighGo CA)
- Below are some of the key members of China PG community that joined the developer Unconferemce :
- Digoal (Alibaba) - Wenjing Zeng (Alibaba) - Yandong Yao (Pivotal) - Zhichun Zhang (Tencent) - Xianwen Zhu (WWIT) - Haibo Lv (Meichuang Technology) - Movead (HighGo)
We have more than 50 participants for the Unconference, the rest of the participants belong to China PostgreSQL community.
- The topics were selected with a voting process from all the participants, around 11 topics were selected for voting. 5 English topics and remaining Mandarin topics. Grant share the English topics on TDE and Sharding shared by Ahsan
- Below were votes casted for the selected topics
| Language | Proposed By | Topic | Number of Votes |
|---|---|---|---|
| English | Ahsan | Specification, design and interface for TDE for PG-14 (Swada-san / Cary / Bruce) | 23 |
| English | Ahsan | Remaining challenges with KMS, finalising the interface and design for PG-14 for KMS. (Swada-san / Bruce) | 19 |
| English | Ahsan | Design for global CSN based snapshot integrated with two phase commit for Postgres FDW (Usama / Andrey) | 24 |
| English | Ahsan | Tooling required for Sharding (This is not a code related topic but it would be interesting to have a discussion on the kind of tooling required Sharding) - Bruce | 18 |
| English | Ahsan | Tooling required for Sharding (This is not a code related topic but it would be interesting to have a discussion on the kind of tooling required Sharding) - Bruce | 18 |
| Mandarin | 朱贤文 | Proposal to replace the current TCP/IP replication protocol by low-latency RDMA. RoCEv2 recommended | 14 |
| Mandarin | 朱贤文 | Currently the database supports 2-phase transaction submission, we want to change to 3-phase transaction submission (mainly for reliability) | 9 |
| Mandarin | 朱贤文 | WAL's parallel recovery function | 10 |
| Mandarin | 朱贤文 | Default setting for shared buffer, refering Oracle Shared Buffers. (Buffer Mapping Partition Lock Setting) | 7 |
| Mandarin | 吕海波 | Increase the HASH table bucket during initialization stage | 9 |
| Mandarin | 吕海波 | make allPgXact struct array with order to reduce the CPU cost during MVCC | 10 |
| English | Cary | Make pgcrypto more secure with key management features (KMS) | 11 |
- It was decided to split the meeting into two zoom rooms, one for English topics and one for Mandarin topics. The max time decided for each topic was 30 minutes.
- After deciding the topics, the next stage of Unconference started after a 10 minutes break.
- Ahsan lead the discussion for English topics with Bruce Momjain, the main topics for English was Sharding and TDE.
- The group discussed the current state of Sharding feature in the community and discussed the issues around using the Clock-SI algorithm for implementing global snapshot feature. Usama pointed out that the community is stuck on couple of issues for global snapshot, the first issue is the patent issue with Clock-SI and second is the issue of linear-ability due to Cock skew.
- Swada pointed out that we can look at other alternatives that involved a coordinator node and worker node, this is similar to the design followed by Postgres-XL and XC. Bruce and Ahsan mentioned that we have run into issues with global transaction with this approach in the past so it is not worth pursuing this one.
- Fujita-san pointed that he is working on the updated patch for FDW async append, he will submit the patch to hackers this week. Bruce pointed out that getting the parallel foreign scan is really important to use the OLAP use-case working for Sharding.
- Usama has taken an action items to send an email to community about the global snapshot linear-ability issue in order to get community opinion
- Bruce has taken an action item to check the Clock-SI patent issue with EDB legal team, Ahsan to share deals with Bruce.
- For the TDE feature, Bruce gave an update on the current status of KMS and TDE in the community. The feature had stalled for a while but Bruce has picked it up. The current target is to get the basic KMS infrastructure and API committed for PG-14, it will generate two keys and won’t do much more.
- Getting the cluster wide TDE would be a target for PG-15, Bruce said that he don’t envision doing user level keys so we will target for cluster level TDE.
- Cary brought up the point of whether the KMS should be part of the Core of external, to this Bruce mentioned that external KMS can be used at boot time.
- Ahsan mentioned that HighGo has an employee that is currently reviewing the TDE work in the community, he will be contributing to TDE in the community.
The English session discussion lasted for an hour, it was very fruitful and productive discussion.
