feat: improved user auth tokens

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!

[mdtro]

Supports getsentry/rfcs#32.

• Newly created user auth tokens will be prefixed with sntryu_.
• Introduce a custom model manager for ApiToken to handle the unique creation logic where we need to hash the token values and store them.
• Use the token_type (currently optional) parameter/field on ApiToken when creating user auth tokens to let the model do the heavy lifting on generating and hashing the values. This will keep the logic out of views and simplify calls to just new_token = ApiToken.objects.create(user=user, token_type=AuthTokenType.USER).
• I've changed some behavior where we only return the refreshToken on applicable token types.

I introduce a "read once" pattern in this PR for the token secrets to prevent leaking of them in logs, exceptions, etc. It works like this when creating a new ApiToken:

1. The model manager sets temporary fields __plaintext_token and __plaintext_refresh_token that store the respective plaintext values for temporary reading.
2. When reading the value through the _plaintext_token property on ApiToken (notice the single prepended underscore) the string value is returned and __plaintext_token is immediately set to None.
3. If you attempt to read the _plaintext_token property again, an exception will be raised, PlaintextSecretAlreadyRead.
   • I opted to raise an exception here rather than just returning None as I feel it's more explicit and will prevent unexpected behavior when we accidentally expect a value to be there.

[codecov]

Codecov Report

Attention: Patch coverage is 97.93814% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 79.68%. Comparing base (7a2578b) to head (42bb344).
Report is 8 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #68148 +/- ##
==========================================
+ Coverage 78.75% 79.68% +0.93% 
==========================================
Files 6428 6429 +1 Lines 285278 285379 +101 Branches 49136 49158 +22 ==========================================
+ Hits 224681 227415 +2734 + Misses 60229 57596 -2633 
Partials 368 368 

Files	Coverage Δ
src/sentry/api/endpoints/api_tokens.py	100.00% <100.00%> (ø)
src/sentry/api/serializers/models/apitoken.py	100.00% <100.00%> (ø)
src/sentry/testutils/factories.py	96.00% <100.00%> (+0.34%)	⬆️
src/sentry/testutils/helpers/backups.py	99.70% <100.00%> (+<0.01%)	⬆️
src/sentry/web/frontend/setup_wizard.py	97.00% <100.00%> (+0.03%)	⬆️
src/sentry/models/apitoken.py	98.96% <97.75%> (-1.04%)	⬇️

... and 234 files with indirect coverage changes

[codecov]

Bundle Report

Changes will increase total bundle size by 12.94kB ⬆️

Bundle name	Size	Change
sentry-webpack-bundle-array-push	26.28MB	12.94kB ⬆️

[nhsiehgit]

hmmm this looks mostly sane to me?

I feel like i would need to significantly refresh my context here to be fully comfortable to sign off on it 😅

[azaslavsky]

Awesome, LGTM from the relocation perspective!

[ykamo001]

Do we want to support these many types? Also it looks like we default to this empty type, would we need the other types and None?

It might be helpful to add some doc strings to this function so people know how to use it and what the intention/difference is based on the input params

[mdtro]

I was running in to some issues with mypy on this when I originally just had the type as AuthTokenType.
The particular field on the class is a CharField so it's expecting a str.

Is there a way around this?

https://github.com/getsentry/sentry/blob/3a86fa61bf4293c0eed9837f6b0529bda15dfbc4/src/sentry/models/apitoken.py#L135

[ykamo001]

It just depends what you want to support in the method, and the caller has to make sure to appease the contract. What is the expected flow or contract we want to have?

[mdtro]

Ultimately, as we improve all of the various token types, I would like to require AuthTokenType. All newly generated tokens would be classified with this and we wouldn't allow a None. Right now, the None is there for backwards compatibility.

Since we've set choices on the CharField, then I suspect we will end up with a validation error if someone provides a string that doesn't match one of the choices. 🤔 I'll test that out as we progress through the token types.

[ykamo001]

Thanks for making all the changes!! Excited about the new flow!