Skip to content

feat(mel)!: the Microsoft.Extensions.Logging integration no longer initializes the SDK - #5595

Merged
jamescrosswell merged 72 commits into
version7from
feat/no-init-from-logging-mel-5245
Oct 1, 2026
Merged

jamescrosswell merged 72 commits into
version7from
feat/no-init-from-logging-mel-5245

Conversation

@jamescrosswell

@jamescrosswell jamescrosswell commented Sep 17, 2026 •

Copy link
Copy Markdown
Collaborator

The Microsoft.Extensions.Logging portion of #5245, stacked on #5592 (log4net) and following the same design. AddSentry now only wires up the logger providers; Sentry has to be initialized separately via SentrySdk.Init, UseSentry, etc.

This is the last of the four logging integrations, so it closes the issue.

Closes #5245

Tip

Background for reviewers on how Sentry gets initialised today, and why this PR splits SentryLoggingOptions: #5595 (comment)

Changelog Entry

  • The Microsoft.Extensions.Logging integration no longer initializes the SDK. Sentry must now be initialized separately from the logging integration (using SentrySdk.Init or UseSentry) - #5595
  • Sentry's logging configuration section (Logging:Sentry) now only configures the logging integration. Core SDK settings there, such as Dsn, no longer have any effect and fail at startup with migration guidance: move them to the Sentry section, or pass them where Sentry is initialized - #5595
  • Blazor WebAssembly's logger now respects the MinimumEventLevel, MinimumBreadcrumbLevel, log entry filters and ConfigureScope callbacks set in UseSentry - #5595

Breaking changes

  • SentryLoggingOptions no longer derives from SentryOptions, matching the Serilog and NLog options. It carries only MinimumBreadcrumbLevel, MinimumEventLevel and log entry filters, so builder.Logging.AddSentry(o => o.Dsn = "…") now fails with a migration error instead of silently doing nothing. Core SDK settings go on the options used to initialize Sentry.
  • SentryLoggingOptions.ConfigureScope is removed. Call SentrySdk.ConfigureScope after initializing Sentry.
  • ILoggingBuilder.AddSentry(string dsn) no longer initializes Sentry. It is kept as an obsolete-as-error tombstone that throws NotSupportedException with migration guidance, so both code callers and reflection-based callers fail loudly.
  • ILoggerFactory.AddSentry(…) no longer initializes Sentry, replaces the current hub, or assigns a MelDiagnosticLogger as the SDK's DiagnosticLogger.
  • InitializeSdk no longer has any effect. SentryLoggingOptions.Dsn and SentryLoggingOptions.InitializeSdk are kept as tombstones that throw when set, including when they are bound from the Sentry configuration section, so an app that upgrades with a config-bound DSN fails at startup instead of silently reporting nothing. On the framework options it is gone entirely.
  • SentryAspNetCoreOptions, SentryMauiOptions and SentryBlazorOptions now derive from a new abstract SentryHostOptions : SentryOptions. They still have MinimumBreadcrumbLevel, MinimumEventLevel, ConfigureScope and AddLogEntryFilter, and the same keys still bind from the Sentry configuration section, so existing UseSentry callbacks and appsettings.json files keep working. Code that treats them as a SentryLoggingOptions no longer compiles.
  • The logging configuration section no longer configures the SDK. Logging:Sentry binds MEL's own LogLevel rules plus MinimumEventLevel and MinimumBreadcrumbLevel; any other key there throws NotSupportedException naming the offending keys. Core SDK settings still bind from the root Sentry section for the integrations that initialize Sentry, which is where every sample and docs guide already puts them.
  • ServiceCollectionExtensions.AddSentry<TOptions> now requires TOptions : SentryHostOptions.
  • builder.Logging.AddSentry() no longer registers SentryOptions in the service collection, so resolving SentryOptions from DI on that path now throws. The MEL integration no longer owns an options object that initializes the SDK. The integrations that do initialize still register it.
  • ConfigureScope callbacks on the framework options now run when Sentry is initialized, rather than when the Sentry logger provider is first created.
  • An AddSentry(o => …) call that only sets logging settings still compiles, and on 6.x it initialized the SDK itself, taking the DSN from SENTRY_DSN or a [Dsn] assembly attribute. The integration therefore warns at runtime — on the first log event that would have become a Sentry event, if Sentry is not initialized but a DSN can still be found, it writes one line to standard error, at most once per logger provider. Microsoft.Extensions.Logging has no self-diagnostics channel, so standard error is the only one available; the shared policy lives in Sentry.Internal.UninitializedSdkWarning, added in feat(serilog)!: the Sentry sink no longer initializes the SDK #5573. SentryLogger.Log also now reads IHub.IsEnabled once per call instead of twice.

Before:

var builder = Host.CreateApplicationBuilder();
builder.Logging.AddSentry("https://key@sentry.io/1");

After:

var builder = Host.CreateApplicationBuilder();

using var sentry = SentrySdk.Init(o => o.Dsn = "https://key@sentry.io/1");

builder.Logging.AddSentry();

Fixes

  • Blazor WebAssembly's logger ignored the logging settings in UseSentry. Blazor registered the plain MEL logger provider, which dependency injection built from a separate, default IOptions<SentryLoggingOptions> rather than the SentryBlazorOptions configured in UseSentry. MinimumEventLevel, MinimumBreadcrumbLevel, log entry filters and ConfigureScope were all silently ignored by the logger; SDK initialization itself was unaffected. The same bug exists on main.
  • ASP.NET Core ran its options setup twice. UseSentry registered SentryAspNetCoreOptionsSetup twice, once with the root Sentry section and once with the logging provider configuration (Logging:Sentry). Different sections, but the same setup class, so everything its Configure adds beyond binding was applied twice: two Kestrel deduplication log filters and two TraceIgnoreStatusCodeTransactionProcessor instances. The setup now binds the Sentry section once, and the logging section gets its own setup that only applies logging settings. On netstandard2.0 the section was bound by a plain Configure<SentryAspNetCoreOptions>(section) call and the setup ran once, so the extras were already applied once there; that branch is gone, and the setup now runs once on every target.
  • Configuration could override Blazor WebAssembly's platform defaults. DetectStartupTime, RequestBodyCompressionLevel and IsGlobalModeEnabled are set because Blazor WASM requires them (System.Diagnostics.Process isn't supported there, and HttpClientHandler can't compress), but they are applied in the same Configure delegate as the UseSentry callback, which was registered before the configuration binding. IConfigureOptions<T> run in registration order, so a Sentry:DetectStartupTime key in wwwroot/appsettings.json won over the platform default, reintroducing the PlatformNotSupportedException it exists to avoid - and configuration won over anything set in UseSentry, the opposite of ASP.NET Core. The setups are now registered first, so precedence runs from the Sentry section, through Logging:Sentry, to the callback and the platform defaults. version7 has the same inversion, where the generic AddSentry<SentryBlazorOptions>(delegate) registered the callback before its options setup.
  • Structured logs from plain MEL took default attributes from the wrong options. This came from the first commit on this branch: sentry.environment, sentry.release and server.address were read from SentryLoggingOptions, which the SDK was no longer initialized with. The structured logger now reads them from the hub.

Notes for review

  • UseSentry no longer loses a TryAdd race with Logging.AddSentry(). The logging integration registers a non-initializing Func<IHub>, so whichever call ran first used to win. Calling builder.Logging.AddSentry() before UseSentry left the SDK disabled with no indication, where v6 failed loudly at startup with "You must supply a DSN". The initializing registration now replaces any existing accessor, so order no longer matters; both orders are tested. Calling both is still redundant and still registers two pairs of logger providers, unchanged from v6. MAUI was never affected, since it initializes in SentryMauiInitializer rather than through the hub accessor.

  • Why SentryHostOptions. SentryLoggingOptions was doing two jobs: configuring the MEL logger, and acting as the base class for integrations that initialize the SDK. Splitting them is what lets SentryLoggingOptions go standalone. The host options pass the log levels and filters through to an inner SentryLoggingOptions instance (the same object, not a copy), and that is what their logger providers receive.

  • Why abstract. It's the natural options type for a generic-host init path (Add a non-logging way to initialise Sentry in generic host apps #5572). Making it concrete later is additive; the reverse would be breaking. It lives in Sentry.Extensions.Logging because that's the one package ASP.NET Core, MAUI and Blazor all reference. Where it ultimately belongs is the packaging question Add a non-logging way to initialise Sentry in generic host apps #5572 raises.

  • ConfigureScope timing. For MAUI and Blazor, the only thing that applied these callbacks used to be the MEL logger provider's constructor, gated on hub.IsEnabled when the provider was built. They're now applied right after init: in AddSentry<TOptions>'s hub factory (ASP.NET Core, Blazor) and in SentryMauiInitializer (MAUI). ASP.NET Core still also applies them per request in SentryMiddleware, and gRPC in its interceptor.

  • SDK name. SentryLoggerProvider no longer stamps Sdk.Name/Sdk.Version, pushes a scope, or disposes the hub. Per Metrics and SentrySdk.Logger logs emitted during a request carry no sentry.sdk.name/sentry.sdk.version on ASP.NET Core #5497 the SDK name identifies the integration that initialized the hub, and the logging integration is identified by the origin (auto.log.extensions_logging). ASP.NET Core, MAUI and gRPC set their own names, so they are unaffected; Blazor WebAssembly has none of its own, which is Blazor WebAssembly apps don't report a Blazor SDK name #5613.

  • Migration guard. Mirrors feat(serilog): configuring a DSN on the sink now fails with a migration error #5611 (Serilog) and the NLog and log4net guards on this stack. A single SentryLoggingConfiguration now binds the logging section on every target framework: it applies the two levels, ignores MEL's own LogLevel rules, and throws for anything else - the obsolete-as-error message for Dsn or InitializeSdk: true, and a message naming the offending keys otherwise. That replaces BindableSentryLoggingOptions and the separate netstandard2.0 reflection path, which is what made the tombstone setters value-dependent (the configuration binder writes each property's own value back, and the unconditional version broke Windows CI earlier on this branch). The setters stay value-dependent, since anyone can still call configuration.Bind(options) themselves.

  • Which configuration each integration reads. The integrations that initialize Sentry read Logging:Sentry from the app's configuration rather than through ILoggerProviderConfiguration, because an app that passes the whole configuration to AddConfiguration - as Sentry.Google.Cloud.Functions does - makes the root Sentry section part of the provider section, and every core setting in it would then be rejected. Plain MEL keeps the provider configuration, where rejecting core settings is right: nothing else binds them onto SentryLoggingOptions, so they really would be ignored.

  • InitializeSdk. MAUI initializes in SentryMauiInitializer, so it calls an internal non-initializing overload of AddSentry<TOptions> instead of setting a flag. Tests that set InitializeSdk = false to avoid initializing now use DisableSdkDsnValue. UseSentry_OptionsNotInitializeSdk_DisabledSdk tested the flag itself and is deleted; UseSentry_DisableDsn_DisabledSdk still covers a disabled SDK.

  • Blazor registrations. UseSentry now delegates to an internal ILoggingBuilder extension so it can be unit tested (WebAssemblyHostBuilder needs a browser runtime). The providers are registered by factory so they keep their existing types, and with them the existing provider alias and filter configuration. Configuration binding moved to a small SentryHostOptionsSetup<TOptions> in Sentry.Extensions.Logging, which has the configuration-binding source generator enabled, since Blazor WASM is trimmed.

  • New tests.

    • The DI init path runs ConfigureScope callbacks, and the non-initializing path doesn't.
    • MAUI ConfigureScope data reaches events. Nothing tested that before.
    • The Blazor logger respects MinimumEventLevel.
    • The Sentry section configures both halves of the host options, the logging section configures only the logging half, a core setting there throws, and the section's extras are applied once.
    • An app that hands its whole configuration to AddConfiguration still gets its root Sentry section bound rather than rejected.
    • Every public setter on SentryLoggingOptions is a key the logging section accepts, so a new logging option can't be mistaken for an SDK setting.
    • Blazor's platform defaults survive configuration keys that would otherwise replace them, and the UseSentry callback wins over configuration while configuration still applies to what the callback leaves alone.

    Each test was checked to fail with its fix reverted.

  • The ApplyDefaultTags tests only used SentryLoggingOptions as a vehicle for a core SentryOptions method; they moved to SentryOptionsTests.

  • The structured-logger tests now give their mocked hub options through SentryOptionsForTestingOnly and reset it afterwards. The MEL provider test previously passed only because another test class happened to leave it set.

  • builder.Logging.AddSentry(dsn) was the only way to initialize Sentry in a generic-host app, so samples/Sentry.Samples.GenericHost now calls SentrySdk.Init directly. Add a non-logging way to initialise Sentry in generic host apps #5572 tracks the replacement and should land before this ships.

  • samples/Sentry.Samples.GenericHost kept its log levels in the root Sentry section and passed the whole configuration to builder.Logging.AddConfiguration, which is how they reached the logging integration. They now live under Logging:Sentry, and that call is gone.

  • SentryWebHostBuilderExtensionsTests' IConfiguration substitute returned "" for every key, where configuration returns null for an absent one. It now returns null, which removed a #if NET8_0-only workaround in UseSentry_Logging_AddLoggerProviders.

  • ApiApprovalTests.Run.Net4_8 can't regenerate on macOS. It was byte-identical to the other snapshots before this change, so it's a copy of the regenerated one.

🤖 Generated with Claude Code

jamescrosswell and others added 10 commits September 14, 2026 12:42
The Sentry sink for Serilog now only configures the sink. Sentry must be
initialized separately (SentrySdk.Init, UseSentry, etc).

- SentrySerilogOptions no longer derives from SentryOptions and only
  carries sink settings; InitializeSdk is removed
- Remove the WriteTo.Sentry(string dsn, ...) overload
- Rename ApplySerilogScopeToEvents() to UseSerilog(), make it idempotent
- The sink logs a one-time diagnostic warning when UseSerilog() was not
  called on the options used to initialize Sentry

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Sentry target for NLog now only configures the target. Sentry must be
initialized separately (SentrySdk.Init, UseSentry, etc).

- SentryNLogOptions no longer derives from SentryOptions and only carries
  target settings; FlushTimeout moves onto it directly
- Remove InitializeSdk, Dsn/DsnLayout, Release/ReleaseLayout,
  Environment/EnvironmentLayout and ShutdownTimeoutSeconds. Events take
  release and environment from the SDK options
- Collapse the AddSentry overloads into
  AddSentry(optionsConfig, targetName); the dsn overloads are removed
- The target no longer routes SDK diagnostics to NLog's InternalLogger

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>
Remove SentryTarget.FlushTimeoutSeconds and SentryNLogOptions.FlushTimeout.
When NLog flushes the target, the hub is now flushed with the FlushTimeout
from the options used to initialize Sentry, since the target no longer
owns the SDK.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Sentry appender for log4net now only sends log events to Sentry.
Sentry must be initialized separately (SentrySdk.Init, UseSentry, etc).

- Remove SentryAppender.Dsn and the lazy SDK initialization on first append
- Remove SentryAppender.Environment; events take the environment from the
  options used to initialize Sentry
- Remove the OnClose override, which only disposed the SDK the appender
  had initialized

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…the SDK

Completes the logging-integration part of #5245. The MEL integration now only
wires up the logger providers; Sentry has to be initialized separately.

Unlike Serilog, NLog and log4net, SentryLoggingOptions keeps deriving from
SentryOptions, because SentryAspNetCoreOptions, SentryMauiOptions and
SentryBlazorOptions derive from it and those integrations do initialize the SDK.
InitializeSdk therefore stays as internal plumbing, now defaulting to false and
opted into by the framework integrations that own it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@codecov

codecov Bot commented Sep 18, 2026 •

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 95.89041% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.07%. Comparing base (af47132) to head (d3a2159).

Files with missing lines Patch % Lines
...or.WebAssembly/WebAssemblyHostBuilderExtensions.cs 88.88% 2 Missing ⚠️
.../Sentry.Extensions.Logging/SentryLoggingOptions.cs 75.00% 1 Missing and 1 partial ⚠️
src/Sentry.Extensions.Logging/SentryLogger.cs 93.75% 0 Missing and 1 partial ⚠️
...entry.Extensions.Logging/SentryStructuredLogger.cs 75.00% 0 Missing and 1 partial ⚠️
Additional details and impacted files
@@                            Coverage Diff                             @@
##           feat/no-init-from-logging-log4net-5245    #5595      +/-   ##
==========================================================================
+ Coverage                                   74.96%   75.07%   +0.10%     
==========================================================================
  Files                                         515      519       +4     
  Lines                                       18832    18872      +40     
  Branches                                     3655     3658       +3     
==========================================================================
+ Hits                                        14118    14168      +50     
+ Misses                                       3855     3840      -15     
- Partials                                      859      864       +5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

… framework options

SentryLoggingOptions no longer derives from SentryOptions, matching the Serilog
and NLog options: it carries only the log levels and entry filters.
SentryAspNetCoreOptions, SentryMauiOptions and SentryBlazorOptions now derive from
a new abstract SentryHostOptions, which keeps MinimumEventLevel,
MinimumBreadcrumbLevel, ConfigureScope and AddLogEntryFilter by passing them
through to an inner SentryLoggingOptions, so existing UseSentry callbacks and
configuration keys keep working.

InitializeSdk is removed. Integrations that initialise through DI call
AddSentry<TOptions>; MAUI, which initialises in SentryMauiInitializer, uses an
internal non-initialising overload. ConfigureScope callbacks are applied right
after the SDK is initialised instead of when the MEL logger provider is built.

Also fixes Blazor WebAssembly's logger ignoring the logging settings from
UseSentry (it was built from a separate, default IOptions<SentryLoggingOptions>),
and structured logs from plain MEL taking default attributes from options the
SDK was not initialised with.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@jamescrosswell

Copy link
Copy Markdown
Collaborator Author

Some background for reviewers on why this PR splits the options class.

Initialising Sentry

Generally Sentry's integrations fall into two camps:

  1. Integrations for features (e.g. databases)
  2. Integrations for platforms (e.g. ASP.NET Core or MAUI)

As a rule, we provide UseSentry extensions on the Builders for the platforms that let you initialise the Sentry SDK at the same time as configuring the Sentry integration for those platforms. We pass a Sentry DSN into the UseSentry method which is used to initialise a Sentry Hub and Client that will be used to send captured information (errors, traces etc.) to the SDK user's Sentry project... and typically there are a bunch of other options specific to that platform that can be set at the same time.

For people not using any specific platform, they can instead initialise Sentry by calling SentrySdk.Init.

One of the features that Sentry supports is logging and it's such a common feature that we decided it might be convenient if people could initialise both logging and the Sentry SDK in a single call (e.g. when using Sentry in a console application, where you also have logging).

In theory that was fine. In practice it's caused huge headaches (see #5245). If you use Serilog and ASP.NET Core, you have to tell Serilog "send logs, but don't initialise — ASP.NET Core already did that". Users constantly get this wrong, and so did we. So in v7, we're removing the ability to initialise the SDK from the logging integrations. These will now be initialised the same as all the other 'feature' integrations we do.

The old way

In version 6 the options classes look something like this:

classDiagram
    SentryOptions <|-- SentryLoggingOptions
    SentryLoggingOptions <|-- SentryAspNetCoreOptions
    SentryLoggingOptions <|-- SentryMauiOptions
    SentryLoggingOptions <|-- SentryBlazorOptions
Loading

SentryLoggingOptions in that diagram does two things: the options type for the MEL integration, and the base class for the ASP.NET Core, MAUI and Blazor options. Those integrations genuinely need to initialise the SDK so they still need an options class that contains a DSN. But in version 7.0 we don't want the options class used to initialise the MEL integration to have a DSN on it since people shouldn't be initialising the Sentry SDK via that integration anymore.

The new way

As such, in version 7.0 we're splitting that options class up:

  • SentryLoggingOptions — now standalone, covering only "which log entries go to Sentry" (two minimum levels plus filters). It no longer inherits SentryOptions, so you cannot set a DSN on it. This is the options class that will be used to initialise MEL in v7.
  • SentryHostOptions (src/Sentry.Extensions.Logging/SentryHostOptions.cs, the new file) — inherits SentryOptions, so it has the DSN, release, sampling and so on, and is the base for integrations that own initialisation. It also exposes the log-level settings, forwarding them to a SentryLoggingOptions it holds internally, so existing UseSentry(o => o.MinimumEventLevel = ...) code and appsettings.json keep working unchanged... we avoid breaking people's existing code.
classDiagram
    SentryOptions <|-- SentryHostOptions
    SentryHostOptions <|-- SentryAspNetCoreOptions
    SentryHostOptions <|-- SentryMauiOptions
    SentryHostOptions <|-- SentryBlazorOptions
    SentryHostOptions *-- SentryLoggingOptions : holds internally
Loading

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just a note on why SentryHostOptions.Logging is a composite property:

  1. SentryLoggingOptions must be a concrete public class. It's what users configure in builder.Logging.AddSentry(o => …), and when using MEL IOptions<SentryLoggingOptions> resolves from DI for the options binding. IOptions<T> is constrained to class, new(), so this can't be an interface.
  2. SentryHostOptions has to be a SentryOptions. It's the object handed to SentrySdk.InitHub... so it can't descend from SentryLoggingOptions - that's kind of the whole point of this PR stack - removing the DSN (and options that are used for SDK initialisation) from the options that are used to initialise the logging integrations.

jamescrosswell and others added 2 commits September 22, 2026 09:30
The sample sets the DSN in code via UseSentry, so the commented-out Dsn
entry is misleading. EnableTracing is declared on BindableSentryOptions
but never applied, so setting it has no effect.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Emit can run concurrently, so the check-then-set on the warned flag could
let more than one thread log the warning.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Comment thread src/Sentry.Extensions.Logging/SentryLoggerProvider.cs Outdated
jamescrosswell and others added 9 commits September 22, 2026 17:08
…sposes the hub

Sdk.Name and Sdk.Version should identify the integration that initialised the
hub, which after this PR can no longer be a logging integration. The logging
integration identifies itself through the log origin (auto.log.*) instead.
See #5497.

With the SDK name gone, and ConfigureScope callbacks now applied at init, the
scope the provider pushed has nothing left to hold, so it goes too.

Disposing the hub goes as well: whoever initialises the hub owns it, and the
provider is now always handed HubAdapter, which is not IDisposable. The one
fixture that handed it a real Hub now disposes the Hub it created.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The sink identifies itself
through the log origin (auto.log.serilog) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.serilog, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The target identifies itself
through the log origin (auto.log.nlog) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.nlog, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry. With no remaining callers, Constants is deleted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The appender identifies
itself through the log origin (auto.log.log4net) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.log4net, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… set the SDK name

Completes the change across the four logging integrations: the SDK name on a log
should identify the integration that initialised the hub, and the logging
integration identifies itself through the origin (auto.log.extensions_logging).
See #5497.

The ASP.NET Core and MAUI structured logger providers keep passing their own SDK
version: those integrations do initialise the SDK, so the name is theirs to set.

With no remaining callers, Constants and SentryLoggerProvider.NameAndVersion are
deleted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@jamescrosswell

Copy link
Copy Markdown
Collaborator Author

The hub plumbing on the logging path is temporary

Worth recording why this PR fixes the TryAdd race rather than removing its cause.

Logging.AddSentry() still registers more than logging: Func<IHub>, IHub, ISentryClient and the HttpClientFactory instrumentation (SentryHttpMessageHandlerBuilderFilter). That's a leftover from when the logging integration initialised the SDK. Because both paths registered the hub accessor with TryAdd, whichever ran first won, so calling builder.Logging.AddSentry() before UseSentry left the SDK silently disabled — where v6 at least failed loudly with "You must supply a DSN".

This PR makes the initializing registration replace any existing accessor, so the order of the two calls no longer matters. That's deliberately targeted, not the end state.

The end state is #5572: the host-level init path takes ownership of the hub plumbing, ISentryClient and the HttpClient instrumentation, and Logging.AddSentry() registers only its two ILoggerProviders — which can take HubAdapter.Instance directly, the same "never capture a hub, always dereference the current one" contract Func<IHub> provides today (#157). At that point nothing competes for the accessor and the RemoveAll here becomes belt-and-braces.

It can't be done in this PR: builder.Logging.AddSentry() is currently the only thing registering those services for generic-host apps, so removing them before #5572 exists would take the capability away with nothing to move to. #5572 has been updated with that scope, and #5645 covers the regression tests for the multiple-container-build behaviour (#103, fixed by #157) that this plumbing defends and that nothing currently pins — worth landing before the restructuring.

Same reasoning applies to SentryHostOptions living in Sentry.Extensions.Logging in this PR: it's the one package ASP.NET Core, MAUI and Blazor all reference. Where it ultimately belongs is part of #5572's packaging question.

@ric-oliv ric-oliv left a comment •

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I ran the test projects this touches, and the inline comments cover what I found. Events from the logging integration also lose nuget:Sentry.Extensions.Logging in sdk.packages, which #5649 tracks.

Up to you if we include these changes, otherwise looks great! ✅

Comment thread src/Sentry.Extensions.Logging/BindableSentryLoggingOptions.cs Outdated
Comment thread test/Sentry.AspNetCore.Tests/IntegrationMockedBackgroundWorker.cs Outdated
Comment thread test/Sentry.AspNetCore.Tests/IntegrationMockedBackgroundWorker.cs Outdated
ric-oliv and others added 7 commits October 1, 2026 11:06
…5651)

The V3 test project links the other test files added in #5592, but not
SentryAppenderConfigurationBindingTests. That file didn't compile against log4net 3, which
marks the XmlElement parameter of XmlConfigurator.Configure as non-nullable. With a
null-forgiving operator on DocumentElement it compiles, so the V3 project now links it and the
Dsn tombstone behavior is pinned on both log4net versions.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(nlog): Report a stale dsn and accept initializeSdk=false

Both tombstones could leave an upgraded app worse off than it needed to be.

A dsn left in NLog.config was silent with NLog's default settings. NLog swallows the setter's
exception unless throwConfigExceptions is on, so the target attached, Sentry was never
initialized and nothing was printed. The runtime warning also stayed quiet, because it only looks
for a DSN in the environment or an assembly attribute. The Dsn setter now writes the migration
message to standard error before it throws.

initializeSdk="false" was the recommended v6 setting next to UseSentry, and it already matches
the new behavior. With throwConfigExceptions on, it still threw, NLog rejected the whole
configuration and the app lost every NLog target. The setter now only throws for true, like the
Microsoft.Extensions.Logging tombstone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(nlog): report a stale initializeSdk="true" as well

The stale dsn message only came from the Dsn setter, so a config carrying
initializeSdk="true" instead was still silent with NLog's default
throwConfigExceptions: the setter threw, NLog discarded it, the target attached and
nothing was printed. Both setters now go through one report-and-throw helper.

Reporting from both setters means a v6 config carrying dsn and initializeSdk="true"
together would print the same message twice, so the helper reports at most once per
target. A configuration reload builds a new target, and so reports again.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>
…245' into feat/no-init-from-logging-log4net-5245

Picks up #5652: a stale dsn or initializeSdk="true" in NLog.config is now reported to
standard error, and initializeSdk="false" is accepted rather than failing the configuration.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…t-5245' into feat/no-init-from-logging-log4net-5245

Picks up #5651, which runs the appender configuration binding tests against log4net 3 as
well. It adds its own <Compile Include> to Sentry.Log4Net.V3.Tests beside the one for the
uninitialized-SDK tests, so the two merged cleanly.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…t-5245' into feat/no-init-from-logging-mel-5245

Picks up #5652 (a stale dsn or initializeSdk="true" in NLog.config is reported to standard
error, and initializeSdk="false" is accepted rather than failing the configuration) and #5651
(the appender configuration binding tests also run against log4net 3), via the log4net branch.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
SentryConstants.DisableSdkDsnValue is an empty string, and SettingLocator.GetDsn
treats an empty Dsn as unset and falls back to SENTRY_DSN. So on a machine with
that variable set, these two tests initialised a working SDK and captured the
event they assert is never captured. FakeSettings() swaps in a locator that reads
only from its own dictionary, so the fallback can't reach the real environment.

Both tests now pass with and without SENTRY_DSN set. The fallback itself predates
this PR and is also on main: options.Dsn = "" does not disable the SDK when
SENTRY_DSN is set.

Reported by @ric-oliv on #5595.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The logging integrations no longer initialize the SDK, so the logging provider
section is no longer a place where a DSN or any other core SDK setting has an
effect. Reading them there would silently ignore them.

The root 'Sentry' section now configures the whole of SentryHostOptions, and
'Logging:Sentry' configures only the logging half. Any other key there throws,
naming the offending keys, so an app that upgrades with core settings in the
logging section fails at startup instead of reporting nothing.

A single SentryLoggingConfiguration binds the logging section for both the
standalone MEL options and the host options, on every target framework. That
replaces BindableSentryLoggingOptions and the separate netstandard2.0
reflection path.

The integrations that initialize Sentry read 'Logging:Sentry' from the app's
configuration rather than through ILoggerProviderConfiguration: an app that
passes the whole configuration to AddConfiguration, as Google Cloud Functions
does, makes the root 'Sentry' section part of the provider section, and every
core setting in it would then be rejected. Plain MEL keeps the provider
configuration, where rejecting core settings is correct.

Also fixes ASP.NET Core registering SentryAspNetCoreOptionsSetup twice - once
with the 'Sentry' section and once with the logging provider configuration -
which applied everything its Configure adds beyond binding twice: two Kestrel
deduplication log filters and two TraceIgnoreStatusCodeTransactionProcessor
instances.

The IConfiguration substitute in SentryWebHostBuilderExtensionsTests returned
an empty string for every key, where configuration returns null for an absent
one. Returning null removes a NET8_0-only workaround.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

@cursor cursor Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Fix All in Cursor

❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.

Reviewed by Cursor Bugbot for commit 04a5a54. Configure here.

Comment thread src/Sentry.AspNetCore.Blazor.WebAssembly/WebAssemblyHostBuilderExtensions.cs Outdated
jamescrosswell and others added 2 commits October 1, 2026 13:55
Blazor WebAssembly sets DetectStartupTime, RequestBodyCompressionLevel and
IsGlobalModeEnabled because the platform requires them, in the same Configure
delegate that runs the UseSentry callback. That delegate was registered before
the configuration binding, and IConfigureOptions<T> run in registration order,
so a Sentry:DetectStartupTime key in wwwroot/appsettings.json won over the
platform default - reintroducing the PlatformNotSupportedException the default
is there to avoid - and configuration won over anything set in UseSentry, the
opposite of ASP.NET Core.

The two setups are now registered first, so precedence runs from the Sentry
section, through Logging:Sentry, to the callback and the platform defaults.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

@ric-oliv ric-oliv left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Very nice!

Base automatically changed from feat/no-init-from-logging-log4net-5245 to version7 October 1, 2026 21:09
@jamescrosswell
jamescrosswell merged commit 378d9e4 into version7 Oct 1, 2026
40 checks passed
@jamescrosswell
jamescrosswell deleted the feat/no-init-from-logging-mel-5245 branch October 1, 2026 21:09
ric-oliv added a commit that referenced this pull request Oct 5, 2026
Brings 6.12.0 and main's changes since 4af9645 into version7, including
#5640, #5643, #5646, #5656, #5659, #5661, #5662 and #5667.

Conflict resolutions:
- Directory.Build.props: keep version7's 7.0.0 / prerelease.
- LoggingBuilderExtensions.cs: keep version7's side. #5595 removed the
  generic AddSentry<TOptions> that #5640 changed; version7 fixes the same
  Blazor bug in AddSentryBlazor.
- LoggingBuilderExtensionsTests.cs: drop #5640's two
  AddSentry_DerivedOptions_* tests (merged cleanly, don't compile on
  version7).
- ServiceCollectionExtensions.cs: keep version7's side and move #5646's
  issue 103 comment to the registrations in AddSentryHub.
- ServiceCollectionExtensionsTests.cs: keep version7's tests and port
  #5646's two tests to run on both version7 init paths: the host
  (AddSentry<TestHostOptions>(initializeSdk: true)) and SentrySdk.Init
  plus Logging.AddSentry().
- .github/workflows/build.yml: keep version7's .NET 11 integration-test
  steps with #5662's integration-test 3.4.1 pin and #5667's
  ubuntu-24.04 runners.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ric-oliv added a commit that referenced this pull request Oct 5, 2026
…ings.json (#5658)

* feat: Serilog sink no longer initializes the SDK

The Sentry sink for Serilog now only configures the sink. Sentry must be
initialized separately (SentrySdk.Init, UseSentry, etc).

- SentrySerilogOptions no longer derives from SentryOptions and only
  carries sink settings; InitializeSdk is removed
- Remove the WriteTo.Sentry(string dsn, ...) overload
- Rename ApplySerilogScopeToEvents() to UseSerilog(), make it idempotent
- The sink logs a one-time diagnostic warning when UseSerilog() was not
  called on the options used to initialize Sentry

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Accept API verifier changes

* Tweak comments in the samples

* feat: NLog target no longer initializes the SDK

The Sentry target for NLog now only configures the target. Sentry must be
initialized separately (SentrySdk.Init, UseSentry, etc).

- SentryNLogOptions no longer derives from SentryOptions and only carries
  target settings; FlushTimeout moves onto it directly
- Remove InitializeSdk, Dsn/DsnLayout, Release/ReleaseLayout,
  Environment/EnvironmentLayout and ShutdownTimeoutSeconds. Events take
  release and environment from the SDK options
- Collapse the AddSentry overloads into
  AddSentry(optionsConfig, targetName); the dsn overloads are removed
- The target no longer routes SDK diagnostics to NLog's InternalLogger

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Tweaked wording

Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>

* feat: NLog target flushes using the SDK's FlushTimeout

Remove SentryTarget.FlushTimeoutSeconds and SentryNLogOptions.FlushTimeout.
When NLog flushes the target, the hub is now flushed with the FlushTimeout
from the options used to initialize Sentry, since the target no longer
owns the SDK.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat: log4net appender no longer initializes the SDK

The Sentry appender for log4net now only sends log events to Sentry.
Sentry must be initialized separately (SentrySdk.Init, UseSentry, etc).

- Remove SentryAppender.Dsn and the lazy SDK initialization on first append
- Remove SentryAppender.Environment; events take the environment from the
  options used to initialize Sentry
- Remove the OnClose override, which only disposed the SDK the appender
  had initialized

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat: Microsoft.Extensions.Logging integration no longer initializes the SDK

Completes the logging-integration part of #5245. The MEL integration now only
wires up the logger providers; Sentry has to be initialized separately.

Unlike Serilog, NLog and log4net, SentryLoggingOptions keeps deriving from
SentryOptions, because SentryAspNetCoreOptions, SentryMauiOptions and
SentryBlazorOptions derive from it and those integrations do initialize the SDK.
InitializeSdk therefore stays as internal plumbing, now defaulting to false and
opted into by the framework integrations that own it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat: SentryHostOptions replaces SentryLoggingOptions as the base for framework options

SentryLoggingOptions no longer derives from SentryOptions, matching the Serilog
and NLog options: it carries only the log levels and entry filters.
SentryAspNetCoreOptions, SentryMauiOptions and SentryBlazorOptions now derive from
a new abstract SentryHostOptions, which keeps MinimumEventLevel,
MinimumBreadcrumbLevel, ConfigureScope and AddLogEntryFilter by passing them
through to an inner SentryLoggingOptions, so existing UseSentry callbacks and
configuration keys keep working.

InitializeSdk is removed. Integrations that initialise through DI call
AddSentry<TOptions>; MAUI, which initialises in SentryMauiInitializer, uses an
internal non-initialising overload. ConfigureScope callbacks are applied right
after the SDK is initialised instead of when the MEL logger provider is built.

Also fixes Blazor WebAssembly's logger ignoring the logging settings from
UseSentry (it was built from a separate, default IOptions<SentryLoggingOptions>),
and structured logs from plain MEL taking default attributes from options the
SDK was not initialised with.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* docs: drop unused Sentry settings from the Serilog sample appsettings

The sample sets the DSN in code via UseSentry, so the commented-out Dsn
entry is misleading. EnableTracing is declared on BindableSentryOptions
but never applied, so setting it has no effect.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(serilog): make the UseSerilog warning check atomic

Emit can run concurrently, so the check-then-set on the warned flag could
let more than one thread log the warning.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor: the logging integration no longer stamps the SDK name or disposes the hub

Sdk.Name and Sdk.Version should identify the integration that initialised the
hub, which after this PR can no longer be a logging integration. The logging
integration identifies itself through the log origin (auto.log.*) instead.
See #5497.

With the SDK name gone, and ConfigureScope callbacks now applied at init, the
scope the provider pushed has nothing left to hold, so it goes too.

Disposing the hub goes as well: whoever initialises the hub owns it, and the
provider is now always handed HubAdapter, which is not IDisposable. The one
fixture that handed it a real Hub now disposes the Hub it created.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor: the Serilog sink no longer sets the SDK name

Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The sink identifies itself
through the log origin (auto.log.serilog) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.serilog, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor: the NLog target no longer sets the SDK name

Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The target identifies itself
through the log origin (auto.log.nlog) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.nlog, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry. With no remaining callers, Constants is deleted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor: the log4net appender no longer sets the SDK name

Sdk.Name should identify the integration that initialised the hub, which after
this change can no longer be a logging integration. The appender identifies
itself through the log origin (auto.log.log4net) instead.
See #5497.

Events are no longer stamped with sentry.dotnet.log4net, and structured logs no
longer carry it as sentry.sdk.name; both now report the SDK that initialised
Sentry.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor: structured logs from Microsoft.Extensions.Logging no longer set the SDK name

Completes the change across the four logging integrations: the SDK name on a log
should identify the integration that initialised the hub, and the logging
integration identifies itself through the origin (auto.log.extensions_logging).
See #5497.

The ASP.NET Core and MAUI structured logger providers keep passing their own SDK
version: those integrations do initialise the SDK, so the name is theirs to set.

With no remaining callers, Constants and SentryLoggerProvider.NameAndVersion are
deleted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(serilog): configuring a DSN on the sink now fails with a migration error

Serilog configuration providers bind sink arguments by parameter name, so
removing the dsn-first overload made them drop `dsn` silently: the sink still
binds, Sentry is never initialized, and nothing is reported. Keeping the
overload as an [Obsolete(error: true)] tombstone that throws makes both
Serilog.Settings.Configuration (appsettings.json) and Serilog.Settings.AppSettings
(app.config) fail loudly with migration guidance, while code callers get a
compile error instead of a type mismatch on the second argument.

The overload mirrors the surviving overload's parameters plus `dsn`. With only
`string dsn` it loses Serilog's overload ranking whenever a configuration
supplies two or more of the surviving arguments, which would restore the silent
behaviour.

Part of #5245

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* test(serilog): pin the DSN tombstone against Serilog.Settings.Configuration

The migration guard works only because of Serilog's overload ranking, and
nothing exercised that path. These tests bind a sink from IConfiguration
the way a provider does, so a Serilog change that stops selecting the
tombstone fails here rather than silently dropping the DSN again.

Verified they fail without the tombstone overload. Selection behaves the
same on Serilog.Settings.Configuration 3.4.0 (Serilog 2.12) and 10.0.1
(Serilog 4.3); 3.4.0 is referenced to avoid bumping Serilog in the tests.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(nlog): configuring a DSN on the target now fails with a migration error

Mirrors the Serilog guard (#5611). The v6 AddSentry(dsn, ...) overloads and
the SentryTarget.Dsn / InitializeSdk properties come back as tombstones:
obsolete-as-error for code callers, throwing NotSupportedException so
NLog.config bindings fail loudly with migration guidance instead of
reporting an unknown property.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* docs(serilog): reword the DSN migration error

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* docs(nlog): reword the DSN migration error to match Serilog

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(log4net): configuring a DSN on the appender now reports a migration error

Brings back SentryAppender.Dsn as a tombstone: [Obsolete(error: true)] with
a setter that throws NotSupportedException carrying migration guidance, so
code callers get a compile error and XML configs report the message instead
of log4net's "Cannot find Property [Dsn]".

Unlike Serilog and NLog, this cannot fail configuration loading: log4net
catches exceptions thrown while setting a parameter, so the appender is
still attached and the message surfaces through log4net's internal logging.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(mel): configuring a DSN through the logging integration now fails with a migration error

Mirrors the Serilog (#5611), NLog and log4net guards. The v6 AddSentry(dsn)
overload and the SentryLoggingOptions.Dsn / InitializeSdk properties come back as
tombstones: obsolete-as-error for code callers, throwing NotSupportedException so
configuration fails loudly with migration guidance instead of being ignored.

Both binding paths are covered. On .NET 6 and later the Sentry section binds
through BindableSentryLoggingOptions, which now carries these keys and throws
when either is present; on netstandard2.0 the configuration binder sets the
properties directly and the setters throw.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(mel): only treat a supplied Dsn or InitializeSdk as a migration error

The tombstone setters threw unconditionally, which broke every bind of
SentryLoggingOptions on netstandard2.0: ConfigurationBinder reads each property
and writes the value back, so InitializeSdk's own `false` tripped the guard even
when the key was absent. That failed four tests on net48, three of them
pre-existing.

The setters now throw only for a value that asks for something the integration
can no longer do, and BindableSentryLoggingOptions matches, so both binding paths
behave the same: a Dsn or InitializeSdk=true is an error, InitializeSdk=false is
accepted because not initializing is what now always happens.

Covered by a test that binds onto the options directly, which reproduces the
netstandard2.0 write-back on every target framework.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(serilog): the Sentry sink registers the Serilog scope event processor automatically (#5612)

* fix: make the SentryOptions processor collections thread safe

SentryClient enumerates these collections lazily for the whole duration of a capture, and
AddEventProcessor is documented as supporting registration after the SDK is initialised.
They were plain Lists, so appending to one while a capture was in flight threw
InvalidOperationException - which the SDK catches and logs at Debug, silently dropping the
event.

Swap them for ConcurrentBagLite, which snapshots on enumeration. Scope.EventProcessors
already uses it for the same reason.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* fix(serilog): register the Serilog scope event processor automatically

The sink no longer initialises the SDK, so integrators have to call UseSerilog() on the
options used to initialise Sentry. Forgetting it was only reported as a warning gated behind
Debug and DiagnosticLevel, so in practice it was silent.

The sink now registers SerilogScopeEventProcessor itself: at construction when Sentry is
already initialised, otherwise on the first log event. The sink and the processor live in the
same assembly, so no reflection is needed and this stays AOT safe. UseSerilog() is still the
better option - it applies from the first event rather than from the first log line - and the
warning now says so.

Also fixes a feedback loop this exposed. Emit answered a reentrant log event with another
diagnostic, which Serilog routed straight back into the sink, each message embedding the
last. With DiagnosticLevel at Info that produced 55 MB of logs in 17 seconds and the app
stopped serving requests. The SDK-namespace filter that breaks the cycle now runs before the
reentrancy check instead of after it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Removed unnecessary comments

Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>

* fix(serilog): register the scope event processor atomically

Sinks sharing one set of SentryOptions can reach registration concurrently -
each sink's guard is per-instance - so the check and the add have to happen
under a lock, not as check-then-act. The sink now learns from the result
whether it was the one that registered, which is what the warning reports.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor(serilog): use the Lock shim for the registration lock

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Tidy comments

Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>

* docs: samples are exempt from the no-comments rule

Restores the DSN comment dropped from the Serilog sample's appsettings.json,
pointing at where this sample actually sets it, and records in AGENTS.md that
"prefer no comments" covers the library rather than samples - including their
JSON configuration files.

Part of #5245

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Apply suggestion from @jamescrosswell

* feat(serilog): warn at runtime when the sink drops events because Sentry is not initialized

The tombstoned overloads catch everyone who passes a DSN to the sink, but they cannot see
the `WriteTo.Sentry(o => ...)` callback that only sets sink options and gets its DSN from
SENTRY_DSN or a [Dsn] assembly attribute. On 6.x that overload initialized the SDK itself;
now it compiles, nothing calls Init, and the sink drops everything silently.

Warn once, on the first event at or above MinimumEventLevel, when the hub is disabled and a
DSN can still be found. There is no DiagnosticLogger to write to in that state, so the
warning goes to Serilog's SelfLog and to standard error.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(nlog): warn at runtime when the target drops events because Sentry is not initialized

Mirrors the Serilog sink: the tombstoned Dsn/InitializeSdk properties cannot see an
AddSentry(o => ...) call that only sets target options and gets its DSN from SENTRY_DSN or a
[Dsn] assembly attribute, so warn once on the first event at or above MinimumEventLevel when
the hub is disabled and a DSN can still be found. The warning goes to NLog's InternalLogger
and to standard error.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(log4net): warn at runtime when the appender drops events because Sentry is not initialized

Mirrors the Serilog sink and the NLog target: the tombstoned Dsn property cannot see an
appender that is configured with only appender settings and gets its DSN from SENTRY_DSN or a
[Dsn] assembly attribute, so warn once on the first event that would have become a Sentry
event when the hub is disabled and a DSN can still be found. The warning goes to log4net's
LogLog and to standard error.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(mel): warn at runtime when the integration drops events because Sentry is not initialized

Completes the cascade from Serilog, NLog and log4net: the tombstoned Dsn/InitializeSdk
properties cannot see an AddSentry(o => ...) call that only sets logging options and gets its
DSN from SENTRY_DSN or a [Dsn] assembly attribute, so warn once on the first log event that
would have become a Sentry event when the hub is disabled and a DSN can still be found.

Microsoft.Extensions.Logging has no self-diagnostics channel, so the warning only goes to
standard error. The warning is owned by SentryLoggerProvider so that it is shared by every
category's logger rather than repeated per category.

SentryLogger.Log now reads IHub.IsEnabled once per call instead of twice, via the level-only
IsEnabledForLevel.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* docs(mel): match the other integrations' migration-error phrasing

The MEL tombstone said "(or an integration such as UseSentry)" where Serilog, NLog and
log4net all say "(or UseSentry via one of the integrations)". Same wording everywhere now.

Only Sentry.Extensions.Logging's API snapshots carry the message, since the tombstoned
members are on SentryLoggingOptions. Net4_8 can't regenerate on macOS, so it got the same
substitution by hand.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix: UseSentry now initializes Sentry even when Logging.AddSentry ran first

The logging integration registers a non-initializing Func<IHub>, and the
initializing one was registered with TryAdd, so whichever ran first won. Calling
builder.Logging.AddSentry() before UseSentry therefore left the SDK disabled with
no indication, where v6 failed loudly at startup with "You must supply a DSN".

The initializing registration now replaces any existing accessor, so the order of
the two calls no longer matters. Tests cover both orders; the AddSentry-first one
fails without this change.

Calling both is still redundant and still registers two pairs of logger
providers, which is unchanged from v6.

Reported by Cursor Bugbot on #5595.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* test(log4net): Run the configuration binding tests against log4net 3 (#5651)

The V3 test project links the other test files added in #5592, but not
SentryAppenderConfigurationBindingTests. That file didn't compile against log4net 3, which
marks the XmlElement parameter of XmlConfigurator.Configure as non-nullable. With a
null-forgiving operator on DocumentElement it compiles, so the V3 project now links it and the
Dsn tombstone behavior is pinned on both log4net versions.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(nlog): report a stale dsn and accept initializeSdk=false (#5652)

* fix(nlog): Report a stale dsn and accept initializeSdk=false

Both tombstones could leave an upgraded app worse off than it needed to be.

A dsn left in NLog.config was silent with NLog's default settings. NLog swallows the setter's
exception unless throwConfigExceptions is on, so the target attached, Sentry was never
initialized and nothing was printed. The runtime warning also stayed quiet, because it only looks
for a DSN in the environment or an assembly attribute. The Dsn setter now writes the migration
message to standard error before it throws.

initializeSdk="false" was the recommended v6 setting next to UseSentry, and it already matches
the new behavior. With throwConfigExceptions on, it still threw, NLog rejected the whole
configuration and the app lost every NLog target. The setter now only throws for true, like the
Microsoft.Extensions.Logging tombstone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(nlog): report a stale initializeSdk="true" as well

The stale dsn message only came from the Dsn setter, so a config carrying
initializeSdk="true" instead was still silent with NLog's default
throwConfigExceptions: the setter threw, NLog discarded it, the target attached and
nothing was printed. Both setters now go through one report-and-throw helper.

Reporting from both setters means a v6 config carrying dsn and initializeSdk="true"
together would print the same message twice, so the helper reports at most once per
target. A configuration reload builds a new target, and so reports again.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>

* test: keep the real environment out of the disabled-SDK tests

SentryConstants.DisableSdkDsnValue is an empty string, and SettingLocator.GetDsn
treats an empty Dsn as unset and falls back to SENTRY_DSN. So on a machine with
that variable set, these two tests initialised a working SDK and captured the
event they assert is never captured. FakeSettings() swaps in a locator that reads
only from its own dictionary, so the fallback can't reach the real environment.

Both tests now pass with and without SENTRY_DSN set. The fallback itself predates
this PR and is also on main: options.Dsn = "" does not disable the SDK when
SENTRY_DSN is set.

Reported by @ric-oliv on #5595.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat: read core SDK settings only from the Sentry configuration section

The logging integrations no longer initialize the SDK, so the logging provider
section is no longer a place where a DSN or any other core SDK setting has an
effect. Reading them there would silently ignore them.

The root 'Sentry' section now configures the whole of SentryHostOptions, and
'Logging:Sentry' configures only the logging half. Any other key there throws,
naming the offending keys, so an app that upgrades with core settings in the
logging section fails at startup instead of reporting nothing.

A single SentryLoggingConfiguration binds the logging section for both the
standalone MEL options and the host options, on every target framework. That
replaces BindableSentryLoggingOptions and the separate netstandard2.0
reflection path.

The integrations that initialize Sentry read 'Logging:Sentry' from the app's
configuration rather than through ILoggerProviderConfiguration: an app that
passes the whole configuration to AddConfiguration, as Google Cloud Functions
does, makes the root 'Sentry' section part of the provider section, and every
core setting in it would then be rejected. Plain MEL keeps the provider
configuration, where rejecting core settings is correct.

Also fixes ASP.NET Core registering SentryAspNetCoreOptionsSetup twice - once
with the 'Sentry' section and once with the logging provider configuration -
which applied everything its Configure adds beyond binding twice: two Kestrel
deduplication log filters and two TraceIgnoreStatusCodeTransactionProcessor
instances.

The IConfiguration substitute in SentryWebHostBuilderExtensionsTests returned
an empty string for every key, where configuration returns null for an absent
one. Returning null removes a NET8_0-only workaround.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(blazor): bind configuration before the UseSentry callback

Blazor WebAssembly sets DetectStartupTime, RequestBodyCompressionLevel and
IsGlobalModeEnabled because the platform requires them, in the same Configure
delegate that runs the UseSentry callback. That delegate was registered before
the configuration binding, and IConfigureOptions<T> run in registration order,
so a Sentry:DetectStartupTime key in wwwroot/appsettings.json won over the
platform default - reintroducing the PlatformNotSupportedException the default
is there to avoid - and configuration won over anything set in UseSentry, the
opposite of ASP.NET Core.

The two setups are now registered first, so precedence runs from the Sentry
section, through Logging:Sentry, to the callback and the platform defaults.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Format code

* fix(maui)!: options set in UseSentry now take precedence over appsettings.json

SentryMauiOptionsSetup both bound the Sentry configuration section and applied
the settings MAUI forces (global mode, the debug logger, the screenshot and SDK
event processors). It was registered after the UseSentry callback, and
IConfigureOptions<T> run in registration order, so a key in appsettings.json
overrode the same setting passed to UseSentry - the reverse of ASP.NET Core.

Moving the whole setup ahead of the callback would have broken the forced half,
which reads Debug and AttachScreenshot as the callback leaves them. The binding
now lives in SentryMauiConfigurationOptionsSetup, registered before the
callback; SentryMauiOptionsSetup keeps the forced settings and still runs last.

Fixes #5657

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Sentry Github Bot <bot+github-bot@sentry.io>
Co-authored-by: Ricardo Colombo Oliveira <github@ricoliv.com>
ric-oliv added a commit that referenced this pull request Oct 6, 2026
…ment

EnableLogs belongs in the Sentry section. Since #5595, setting it under
Logging:Sentry fails at startup, so the old hint sent readers to the
wrong place.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jamescrosswell added a commit that referenced this pull request Oct 7, 2026
…Blazor, where logs are opt-in (#5642)

* feat(logging): Add EnableLogs to SentryHostOptions

ASP.NET Core, Google Cloud Functions, MAUI and Blazor WebAssembly add
the MEL integration on the user's behalf, and that isn't consent to send
logs. Since #5504, these apps sent every ILogger entry to Sentry.

SentryHostOptions now has EnableLogs, default false. It hides the
obsolete SentryOptions.EnableLogs and binds from the existing EnableLogs
configuration key. SentryStructuredLogger reads it from the hub's
options, so Logging.AddSentry() on a hub created by SentrySdk.Init keeps
sending logs, like the other logging integrations.

With no log envelopes sent by default, the WebIntegrationTests.Versioning
snapshots and the Google Cloud Functions SDK-name assertion return to
how they were before #5504.

Refs #5183

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* chore(samples): Set EnableLogs again in the framework samples

Restores EnableLogs = true where #5504 removed it, so these samples keep
sending logs: ASP.NET Core Basic, gRPC, MVC, ME.AI and MAUI. The Google
Cloud Functions sample already has it again after main was merged into
version7.

Refs #5183

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Accept API verifier changes

* feat(logging)!: Remove SentryOptions.EnableLogs

Only the integrations that derive from Sentry.Extensions.Logging (ASP.NET
Core, MAUI, Blazor WebAssembly) need the option, so it now lives on
SentryHostOptions alone. SentrySdk.Logger and the logging integrations
always send logs, and the logs spec dropped the global option in 3.0.0.
James Crosswell agreed to remove it in v7.

The obsolete base property also made code that set EnableLogs through a
SentryOptions reference compile, warn that logs are always enabled, and
leave them off on host options. That code now fails to compile instead.

NLog's SentryTarget.EnableLogs stays as an obsolete no-op with its own
message, so existing NLog.config files still load.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(aspnetcore): Check EnableLogs end to end through UseSentry

The existing tests build a Hub directly. These run a real UseSentry host,
which reaches the gate through HubAdapter, and cover the default, the
opt-in in code and the opt-in from the Sentry configuration section.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* chore(samples): Clarify what EnableLogs covers in the ASP.NET Core sample

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* feat(nlog)!: Remove SentryTarget.EnableLogs

The Sentry target always sends logs, so the option was an ignored no-op.
v7 removes SentryOptions.EnableLogs and Serilog's enableLogs parameter,
so the NLog target drops its copy too.

A leftover enableLogs attribute in NLog.config is ignored with NLog's
default settings. With throwConfigExceptions or throwExceptions on, NLog
refuses to load the configuration and names the attribute, so the app
fails on its first run after the upgrade.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Accept API verifier changes

* test: Assert the single log in the EnableLogs opt-in tests

ContainSingle with a predicate passed when other logs were also sent.
Asserting a single log, then its message, also catches extra logs, such
as the SDK's own, and the failure shows the log that arrived.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(nlog): Keep SentryTarget.EnableLogs so v6 configurations still load

Removing the property broke NLog.config files that still set enableLogs.
With throwConfigExceptions on, NLog rejected the whole configuration, so
the app lost every target, not only Sentry. #5652 fixed the same failure
for initializeSdk="false".

This reverts 8b80e53. The property stays obsolete and ignored, with its
own obsolete message, because SentryOptions.ObsoleteEnableLogs is gone.
LoadConfiguration_WithEnableLogs_DoesNotThrow loads both values with
throwConfigExceptions on.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* chore(samples): Drop the Logging section hint from the EnableLogs comment

EnableLogs belongs in the Sentry section. Since #5595, setting it under
Logging:Sentry fails at startup, so the old hint sent readers to the
wrong place.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(logging): Declare EnableLogs on the host options, not SentryHostOptions

EnableLogs is public on SentryAspNetCoreOptions, SentryMauiOptions and
SentryBlazorOptions, the hosts that add the ILogger provider for the
user. Each one forwards to an internal flag on SentryHostOptions, so a
future host that derives from it only gets the setting if it declares
one. Settling this before v7 ships avoids a breaking change later.

SentryStructuredLogger still reads the flag from the hub's options, so
an explicit Logging.AddSentry() in a host app stays off unless
EnableLogs is set. BindableSentryHostOptions keeps binding EnableLogs
from the Sentry section for all three hosts.

The remarks no longer mention SentrySdk.Logger and the logging
integrations.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Sentry Github Bot <bot+github-bot@sentry.io>
Co-authored-by: James Crosswell <jamescrosswell@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

risk: high PR risk score: high

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants