fix(azservicebus): set minimum 60s server-timeout for management operations

[EldertGrootenboer]

Problem

Receiver.PeekMessages() consistently times out with context deadline exceeded on specific queues (#26421), while Receiver.ReceiveMessages() on the same queue succeeds quickly.

Root cause: The Go SDK passes the user's context deadline directly as the server-timeout AMQP application property on management RPC operations. The .NET SDK sends a fixed 60,000 ms and Java sends 59,000 ms. With typical user contexts of 10-20 seconds, the service doesn't have enough time to complete the partition scatter-gather that PeekMessages triggers on partitioned entities.

The service-side PeekRequestAsyncResult iterates all partitions sequentially, forwarding each peek to the partition's broker via an internal RPC link. Each hop uses RemainingTime() as its deadline, so slow partitions cascade into timeout exhaustion. ReceiveMessages bypasses this entirely by using a direct data link to a single partition.

Fix

Added a defaultServerTimeout constant (60 seconds) and serverTimeoutMillis(ctx) helper that returns max(ctx_remaining, 60s). Applied to all management operation enforcement points:

• mgmt.go — PeekMessages, ScheduleMessages, CancelScheduledMessages (explicit)
• rpc.go — RPC fallback covering ReceiveDeferred, RenewLocks, RenewSessionLock, GetSessionState, SetSessionState, SettleOnMgmtLink

The user's context still governs client-side cancellation (the select on ctx.Done() in rpc.go is unchanged). The fix only affects the server-timeout value sent to the service, giving it sufficient headroom for partition scatter-gather.

Design decisions

• 60s floor (not fixed): Unlike .NET's fixed 60s, Go uses max(remaining, 60s). This respects users who set longer timeouts (e.g., 120s) while establishing a floor that matches cross-SDK parity.
• All management ops (not just PeekMessages): All management operations go through the same RPC path and could be affected by partition scatter-gather. Scoping to PeekMessages only would leave others vulnerable.
• Dual-key guard in rpc.go: ScheduleMessages/CancelScheduledMessages use the vendor-prefixed key com.microsoft:server-timeout while PeekMessages and the RPC fallback use server-timeout. The guard now checks both keys to avoid sending a duplicate property.
• No-deadline contexts now send explicit 60s: Previously, no server-timeout was sent when ctx had no deadline. Now 60000 is always sent, matching .NET/Java behavior.

Testing

6 table-driven unit tests for serverTimeoutMillis covering: no deadline, short deadline (10s floor), very short (1s floor), long deadline (120s respected), exact boundary (60s), and expired context. All pass.

Full module suite: 10/10 test packages pass, go vet clean, gofmt clean.

Cross-SDK Parity

SDK	server-timeout	Source
.NET	Fixed 60,000 ms	ServiceBusRetryOptions.cs, AmqpRequestMessage.cs
Java	Fixed 59,000 ms	ServiceBusConstants.java, MessageUtils.java
Go (before)	User's ctx deadline	—
Go (after)	max(ctx_remaining, 60,000) ms	mgmt.go:serverTimeoutMillis()

Fixes #26421

[Copilot]

Pull request overview

This PR adjusts how server-timeout is computed and applied for Service Bus RPC-based operations to reduce premature timeouts (notably Receiver.PeekMessages() on partitioned entities), aligning behavior more closely with other language SDK defaults.

Changes:

• Introduces a defaultServerTimeout (60s) and serverTimeoutMillis(ctx) helper that uses max(ctx_remaining, 60s) (or 60s if no deadline).
• Applies the computed server-timeout to management operations (explicitly in mgmt.go and as an RPC fallback in rpc.go, with a dual-key guard).
• Adds unit tests covering serverTimeoutMillis behavior across deadline/no-deadline scenarios.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
sdk/messaging/azservicebus/internal/rpc.go	Ensures RPC requests include a server-timeout (unless already set), with a guard against duplicate vendor-prefixed keys.
sdk/messaging/azservicebus/internal/mgmt.go	Adds the 60s floor helper and applies it to management calls like Peek/Schedule/Cancel.
sdk/messaging/azservicebus/internal/mgmt_test.go	Adds table-driven unit tests for serverTimeoutMillis.
sdk/messaging/azservicebus/CHANGELOG.md	Documents the behavioral change for management operations’ server-timeout.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated no new comments.