feat: Add common persistence. #193
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Moves the client's `ItemDescriptor` to internal where it can be shared, and gives it a generic parameter so it can hold flag rules/segments.
Augments event processor with context-key deduplication abilities using an LRU cache.
This PR extends our existing deserialization/serialization technique from the client-side SDK to the server-side SDK data model. Specifically, it adds deserialization for the segment structure. I've also introduced some helpers to make the logic less repetitive, and centralize some decisions (like what to do if a JSON field isn't present, or is null.) As background, the `tag_invoke` functions present in this PR are what allows the `boost::json` library to recursively deserialize complex structures. Here, I've added a couple new `tag_invokes` for bool, string, vector, etc. The special part is that they deserialize into `expected<type, error>` rather than the primitive directly; this allows us to return an error if the deserialization failed.
This PR builds on #153, extending the deserialization to flag data. I've added to the `PARSE_FIELD` macro to handle the following cases: 1. Parse a field where the zero-value is a valid part of the domain: `PARSE_FIELD`. In other words, if the field is omitted or null, it will be filled in with a default value. Most common scenario. 2. Like (1), but supply a default value in case the field is null or omitted: `PARSE_FIELD_DEFAULT`. 3. Parse a field that doesn't have a valid default value, and so is represented as an `optional`: `PARSE_CONDITIONAL_FIELD`. For example, the value `""` wouldn't be a valid default for a unique key. 4. Parse a field that must be present in the JSON data: `PARSE_REQUIRED_FIELD`. Usage of this should be rare; the test would be if the SDK is totally unable to function without this. To achieve this, I've made the `tag_invoke`s operate on `expected<optional<T>, JsonError`'s, instead of `expected<T, JsonError`. We could potentially revert that (and instead add a new enum case to `JsonError`), but I figure `optional` is a good representation of null/omitted. I think I need to actually try removing it to determine if it results in a better API. One thing lacking is better error messages - we use kSchemaFailure everywhere but we now have the ability to report exactly what was missing/wrong type, etc.
Adds in missing `Flag` model to the top-level `SDKDataSet`. Additionally, both the flag/segment fields now use use `PARSE_FIELD`, so we can get an empty `unordered_map` if they are missing.
Co-authored-by: Casey Waldren <[email protected]>
…174) Co-authored-by: Casey Waldren <[email protected]>
Co-authored-by: Casey Waldren <[email protected]>
Previously, context kind was an `std::string`. The requirement that it not be an empty string was enforced by parent data model deserializers. Now, we have a dedicated type-safe wrapper via`BOOST_STRONG_TYPEDEF`. The custom `tag_invoke` enforces the invariant so it's not possible to deserialize it inconsistently.
This commit contains the server-side evaluation engine and associated tests.
Adds a server-side Client object.
This significantly refactors the `Variation` internal calls to be much easier to reason about. In the process, it fixes some broken behavior which is now reflected in the suppressions list. It also refactors event generation to use the `EventScope`/`EventFactory` pattern found in other SDKs, such as Rust and Go.
This removes a bunch of duplicated work from the top-level typed Variation methods, and pushes it into the internal `Variation` method.
We had an infinite loop where the `Value` tag invoke deserializer was calling the `tl::expected<Value, JsonError` version, which called back to itself. This was manifesting as a runtime infinite loop. It's not caught by contract tests since I believe the trigger is having a list of variations that are themselves arrays, like: ``` "variations": [["a","b","c",123,true,"false"], [ .. other stuff ..], ... ] ``` The fix is to break the cycle and remove the incorrect implementation.
This moves all the `_ErrorInfo_` C bindings out of the client-side SDK and into the common library, allowing them to be shared with the server. This should be backwards compatible because the `sdk.h` includes the new `error_info.h`, providing the same symbols as before.
Adds a new README for the server. Also adds a link from the client-side README to this one. --------- Co-authored-by: Molly <[email protected]>
I hadn't wired up the `DataStoreUpdater` component, so we were just inserting flag updates directly in the memory store. With this in place, we can remove the streaming update suppression.
Implements a set of C bindings for the server-side SDK. Missing are flag notifier bindings, but this isn't implemented in the C++ side yet either.
Hooks up the initial reconnect delay parameter to the server-side streaming data source. Allows us to remove another batch of contract test failures.
The argument to `ValidChar` is a `char`, so its max value is `255` rendering the check redundant. This causes a warning on our Mac compilations as well as a linter warning. --------- Co-authored-by: Ryan Lamb <[email protected]>
cwaldren-ld
force-pushed
the
rlamb/common-persistent-store
branch
from
bfcb09c to
8345ebb
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.