This is general knowledge or an external-engine reference, not an Apollo architecture or implementation claim. See Apollo's architecture status for target terminology and verified implementation boundaries.
Q31: How Do You Design a Friend System?
Core Conclusion
A friend system looks like a social feature, but it is actually a composite system of "relationship storage plus online presence plus interaction permissions".
What really needs to be thought through first is not interface details, but:
- Whether relationships are bidirectional or unidirectional
- How online state propagates
- How blacklist and permissions are adjudicated
- How relationships are maintained after cross-server and partitioning
If these boundaries are not defined clearly, later features pile up messily.
1. First Define Relationship Types
A friend system usually has more than one kind of relationship.
Common ones include:
- Bidirectional friends
- Unidirectional follows
- Blacklists
- Special relationships, such as intimacy, master/apprentice, CP
The storage models and notification rules of these relationships are all different; they should not be mixed into one fuzzy logic.
2. The Real Core Capabilities of a Friend System
It usually includes at least:
- Establishing and dissolving relationships
- Friend list queries
- Online state viewing
- Interaction entries such as private chat, invites, and teleports
- Blacklist interception
On the surface it is "adding friends"; in essence it is social relationships and permission adjudication.
3. How to Model Relationship Data
The most common approach is a relationship table, rather than stuffing the whole friend list into the character master record.
The reasons are simple:
- Relationships are naturally many-to-many
- They need separate queries and pagination
- They need to carry extra information such as remarks, groups, and intimacy
Bidirectional friends usually have two modeling approaches:
- One record representing a pair of relationships
- Two directional records representing each side's perspective
Which to choose depends on query and maintenance habits.
4. Why Online State Is the Hard Part
The static relationships of a friend system are not hard; what changes frequently is online state.
The difficulty lies in:
- Players frequently going online and offline
- Possibly crossing gateways and servers
- State pushes easily forming fan-out
So many systems do layering:
- Relationship data goes to the persistence layer
- Online state goes to a cache or session layer
- Notifications are by subscription or incremental update
5. Do Not Make the Friend System a "Full Push System"
When the friend count is large, pushing the full state to all friends on every state change is very costly.
A more reasonable approach is usually:
- Pull a summary on login
- Push only incremental changes on online/offline
- Relationships with no interaction for a long time get lower update frequency or lazy queries
6. Blacklist and Permissions Must Have Higher Priority
Many systems treat the blacklist as an add-on feature, which is dangerous.
The more reasonable principles are:
- Blacklist takes priority over friend relationships
- After being blacklisted, entries such as private chat, invites, and applications should be blocked
- Certain display information should also be restricted
Otherwise the logic will conflict with itself.
7. What Needs Extra Consideration in Cross-Server Scenarios
If the game has cross-server, partitioning, or multi-scene clusters, the friend system usually needs to solve:
- Globally unique character identifiers
- Unified online state sources
- Cross-server private chat and invite routing
- Offline messages or reminders
At this point the friend system is no longer just a single-server table structure problem, but a social platform problem.
8. The More Robust Combination in Engineering
A common approach is:
- The persistence layer saves friend relationships and static data such as remarks
- Redis or an online service saves an online state summary
- A social service uniformly handles applications, blacklists, invites, and private chat routing
This way the core game logic is not dragged down too heavily by social state.
9. Common Misconceptions
1. The Friend System Is Just a Relationship Table
Not enough. Online state, permission control, and cross-server routing are often more complex.
2. Querying the Character Process Directly for Online State Is Fine
In a distributed environment this is very heavy, and not conducive to unified aggregation.
3. The Blacklist Is Only Chat Filtering
Usually it is not. It also affects invites, applications, profile viewing, and other entries.
References
- Various online game social systems and online state service practice materials
