This document is a proposal to work towards reducing and limiting table sizes on GitLab.com. We establish a **measurable target** by limiting table size to a certain threshold. This will be used as an indicator to drive database focus and decision making. With GitLab.com growing, we continuously re-evaluate which tables need to be worked on to prevent or otherwise fix violations.
This document is a proposal to work towards reducing and limiting table sizes on GitLab.com. We establish a **measurable target** by limiting table size to a certain threshold. This will be used as an indicator to drive database focus and decision making. With GitLab.com growing, we continuously re-evaluate which tables need to be worked on to prevent or otherwise fix violations.
Note that this is not meant to be a hard rule but rather a strong indication that work needs to be done to break a table apart or otherwise reduce its size.
This is not meant to be a hard rule but rather a strong indication that work needs to be done to break a table apart or otherwise reduce its size.
This is meant to be read in context with the [Database Sharding blueprint](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/64115),
This is meant to be read in context with the [Database Sharding blueprint](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/64115),
which paints the bigger picture. This proposal here is thought to be part of the "debloating step" below, as we aim to reduce storage requirements and improve data modeling. Partitioning is part of the standard tool-belt: where possible, we can already use partitioning as a solution to cut physical table sizes significantly. Both will help to prepare efforts like decomposition (database usage is already optimized) and sharding (database is already partitioned along an identified data access dimension).
which paints the bigger picture. This proposal here is thought to be part of the "debloating step" below, as we aim to reduce storage requirements and improve data modeling. Partitioning is part of the standard tool-belt: where possible, we can already use partitioning as a solution to cut physical table sizes significantly. Both will help to prepare efforts like decomposition (database usage is already optimized) and sharding (database is already partitioned along an identified data access dimension).
...
@@ -124,7 +124,7 @@ In order to maintain and improve operational stability and lessen development bu
...
@@ -124,7 +124,7 @@ In order to maintain and improve operational stability and lessen development bu
1. Indexes are smaller, can be maintained more efficiently and fit better into memory
1. Indexes are smaller, can be maintained more efficiently and fit better into memory
1. Data migrations are easier to reason about, take less time to implement and execute
1. Data migrations are easier to reason about, take less time to implement and execute
Note that this target is *pragmatic*: We understand table sizes depend on feature usage, code changes and other factors - which all change over time. We may not always find solutions where we can tightly limit the size of physical tables once and for all. That is acceptable though and we primarily aim to keep the situation on GitLab.com under control. We adapt our efforts to the situation present on GitLab.com and will re-evaluate frequently.
This target is *pragmatic*: We understand table sizes depend on feature usage, code changes and other factors - which all change over time. We may not always find solutions where we can tightly limit the size of physical tables once and for all. That is acceptable though and we primarily aim to keep the situation on GitLab.com under control. We adapt our efforts to the situation present on GitLab.com and will re-evaluate frequently.
While there are changes we can make that lead to a constant maximum physical table size over time, this doesn't need to be the case necessarily. Consider for example hash partitioniong, which breaks a table down into a static number of partitions. With data growth over time, individual partitions will also grow in size and may eventually reach the threshold size again. We strive to get constant table sizes, but it is acceptable to ship easier solutions that don't have this characteristic but improve the situation for a considerable amount of time.
While there are changes we can make that lead to a constant maximum physical table size over time, this doesn't need to be the case necessarily. Consider for example hash partitioniong, which breaks a table down into a static number of partitions. With data growth over time, individual partitions will also grow in size and may eventually reach the threshold size again. We strive to get constant table sizes, but it is acceptable to ship easier solutions that don't have this characteristic but improve the situation for a considerable amount of time.
@@ -64,7 +64,7 @@ We already use Database Lab from [postgres.ai](https://postgres.ai/), which is a
...
@@ -64,7 +64,7 @@ We already use Database Lab from [postgres.ai](https://postgres.ai/), which is a
Internally, this is based on ZFS and implements a "thin-cloning technology". That is, ZFS snapshots are being used to clone the data and it exposes a full read/write PostgreSQL cluster based on the cloned data. This is called a *thin clone*. It is rather short lived and is going to be destroyed again shortly after we are finished using it.
Internally, this is based on ZFS and implements a "thin-cloning technology". That is, ZFS snapshots are being used to clone the data and it exposes a full read/write PostgreSQL cluster based on the cloned data. This is called a *thin clone*. It is rather short lived and is going to be destroyed again shortly after we are finished using it.
It is important to note that a thin clone is fully read/write. This allows us to execute migrations on top of it.
A thin clone is fully read/write. This allows us to execute migrations on top of it.
Database Lab provides an API we can interact with to manage thin clones. In order to automate the migration and query testing, we add steps to the `gitlab/gitlab-org` CI pipeline. This triggers automation that performs the following steps for a given merge request:
Database Lab provides an API we can interact with to manage thin clones. In order to automate the migration and query testing, we add steps to the `gitlab/gitlab-org` CI pipeline. This triggers automation that performs the following steps for a given merge request:
