MCPA-Level-1 MuleSoft Certified Platform Architect - Level 1 Question and Answers

Explanation

Correct Answer: Increase the size of the CloudHub worker(s)

*****************************************

The key details that we can take out from the given scenario are:

>> API implementation uses a database bulk insert command to submit all the purchase data to a database

>> JDBC driver processes the data into a set of several temporary disk files on the CloudHub worker

>> Sometimes a request fails and the logs show a message indicating an out-of-file-space message

Based on above details:

>> Both auto-scaling options does NOT help because we cannot set auto-scaling rules based on error messages. Auto-scaling rules are kicked-off based on CPU/Memory usages and not due to some given error or disk space issues.

>> Increasing the number of CloudHub workers also does NOT help here because the reason for the failure is not due to performance aspects w.r.t CPU or Memory. It is due to disk-space.

>> Moreover, the API is doing bulk insert to submit the received batch data. Which means, all data is handled by ONE worker only at a time. So, the disk space issue should be tackled on "per worker" basis. Having multiple workers does not help as the batch may still fail on any worker when disk is out of space on that particular worker.

Therefore, the right way to deal this issue and resolve this is to increase the vCore size of the worker so that a new worker with more disk space will be provisioned.

Explanation

Correct Answer: The API consumer creates an API client, which sends API invocations to an API such that they are processed by an API implementation

*****************************************

Terminology:

>> API Client - It is a piece of code or program the is written to invoke an API

>> API Consumer - An owner/entity who owns the API Client. API Consumers write API clients.

>> API - The provider of the API functionality. Typically an API Instance on API Manager where they are managed and operated.

>> API Implementation - The actual piece of code written by API provider where the functionality of the API is implemented. Typically, these are Mule Applications running on Runtime Manager.

Explanation

Correct Answer: The FQDNs are determined by the application name chosen, IRRESPECTIVE of the region

*****************************************

>> When deploying applications to Shared Worker Cloud, the FQDN are always determined by application name chosen.

>> It does NOT matter what region the app is being deployed to.

>> Although it is fact and true that the generated FQDN will have the region included in it (Ex: exp-salesorder-api.au-s1.cloudhub.io), it does NOT mean that the same name can be used when deploying to another CloudHub region.

>> Application name should be universally unique irrespective of Region and Organization and solely determines the FQDN for Shared Load Balancers.

Explanation

Correct Answer: When the response time of API invocations exceeds a threshold

*****************************************

>> Alerts can be setup for all the given options using the default Anypoint Platform functionality

>> However, the question insists on an alert whose conditions depend on the end-to-end request processing of the API implementation.

>> Alert w.r.t "Response Times" is the only one which requires end-to-end request processing of API implementation in order to determine if the threshold is exceeded or not.

Explanation

Correct Answer: SLA-based rate limiting

*****************************************

>> Client Id enforement policy is a "Compliance" related NFR and does not help in maintaining the "Quality of Service (QoS)". It CANNOT and NOT meant for protecting the backend systems from scalability challenges.

>> IP Whitelisting and OAuth 2.0 token enforcement are "Security" related NFRs and again does not help in maintaining the "Quality of Service (QoS)". They CANNOT and are NOT meant for protecting the backend systems from scalability challenges.

Rate Limiting, Rate Limiting-SLA, Throttling, Spike Control are the policies that are "Quality of Service (QOS)" related NFRs and are meant to help in protecting the backend systems from getting overloaded.

https://dzone.com/articles/how-to-secure-apis

Explanation

Correct Answer: When a pragmatic approach with only limited isolation from the backend system is deemed appropriate.

*****************************************

General guidance w.r.t choosing Data Models:

>> If an Enterprise Data Model is in use then the API data model of System APIs should make use of data types from that Enterprise Data Model and the corresponding API implementation should translate between these data types from the Enterprise Data Model and the native data model of the backend system.

>> If no Enterprise Data Model is in use then each System API should be assigned to a Bounded Context, the API data model of System APIs should make use of data types from the corresponding Bounded Context Data Model and the corresponding API implementation should translate between these data types from the Bounded Context Data Model and the native data model of the backend system. In this scenario, the data types in the Bounded Context Data Model are defined purely in terms of their business characteristics and are typically not related to the native data model of the backend system. In other words, the translation effort may be significant.

>> If no Enterprise Data Model is in use, and the definition of a clean Bounded Context Data Model is considered too much effort, then the API data model of System APIs should make use of data types that approximately mirror those from the backend system, same semantics and naming as backend system, lightly sanitized, expose all fields needed for the given System API’s functionality, but not significantly more and making good use of REST conventions.

The latter approach, i.e., exposing in System APIs an API data model that basically mirrors that of the backend system, does not provide satisfactory isolation from backend systems through the System API tier on its own. In particular, it will typically not be possible to "swap out" a backend system without significantly changing all System APIs in front of that backend systemand therefore the API implementations of all Process APIs that depend on those System APIs! This is so because it is not desirable to prolong the life of a previous backend system’s data model in the form of the API data model of System APIs that now front a new backend system. The API data models of System APIs following this approach must therefore change when the backend system is replaced.

On the other hand:

>> It is a very pragmatic approach that adds comparatively little overhead over accessing the backend system directly

>> Isolates API clients from intricacies of the backend system outside the data model (protocol, authentication, connection pooling, network address, …)

>> Allows the usual API policies to be applied to System APIs

>> Makes the API data model for interacting with the backend system explicit and visible, by exposing it in the RAML definitions of the System APIs

>> Further isolation from the backend system data model does occur in the API implementations of the Process API tier

Explanation

Correct Answer:

1. App Developers owns and focuses on Experience Layer APIs

2. LOB IT owns and focuses on Process Layer APIs

3. Central IT owns and focuses on System Layer APIs

References:

https://blogs.mulesoft.com/biz/api/experience-api-ownership/

https://blogs.mulesoft.com/biz/api/process-api-ownership/

https://blogs.mulesoft.com/biz/api/system-api-ownership/

Explanation

Correct Answers:

1. Each modern API must be treated like a product and designed for a particular target audience (for instance mobile app developers)

*****************************************

Bottom of Form

Top of Form

Explanation

Correct Answer: API implementations can run successfully in customer-hosted Mule runtimes, even when they are unable to communicate with the control plane.

*****************************************

>> We CANNOT use Shared Load balancer to load balance APIs on customer hosted runtimes

>> For Hybrid deployment models, the on-premises are first connected to Runtime Manager using Runtime Manager agent. So, the connection is initiated first from On-premises to Runtime Manager. Then all control can be done from Runtime Manager.

>> Anypoint Runtime Manager CANNOT ensure automatic HA. Clusters/Server Groups etc should be configured before hand.

Only TRUE statement in the given choices is, API implementations can run successfully in customer-hosted Mule runtimes, even when they are unable to communicate with the control plane. There are several references below to justify this statement.

References:

https://docs.mulesoft.com/runtime-manager/deployment-strategies#hybrid-deployments

https://help.mulesoft.com/s/article/On-Premise-Runtimes-Disconnected-From-US-Control-Plane-June-18th-2018

https://help.mulesoft.com/s/article/Runtime-Manager-cannot-manage-On-Prem-Applications-and-Servers-from-US-Control-Plane-June-25th-2 019

https://help.mulesoft.com/s/article/On-premise-Runtimes-Appear-Disconnected-in-Runtime-Manager-May-29th-2018

============================

============================

Explanation

Correct Answer: XML over HTTP

*****************************************

>> API-led connectivity and Application Networks urge to have the APIs on HTTP based protocols for building most effective APIs and networks on top of them.

>> The HTTP based APIs allow the platform to apply various varities of policies to address many NFRs

>> The HTTP based APIs also allow to implement many standard and effective implementation patterns that adhere to HTTP based w3c rules.

Bottom of Form

Top of Form

Explanation

Correct Answer: Implement different system APIs named createCustomer, amendCustomer, retrieveCustomer and suspendCustomer as they are modular and has seperation of concerns

*****************************************

>> It is quite normal to have a single API and different Verb + Resource combinations. However, this fits well for an Experience API or a Process API but not a best architecture style for System APIs. So, option with just one customerManagement API is not the best choice here.

>> The option with APIs in createCustomerInCRMZ format is next close choice w.r.t modularization and less maintenance but the naming of APIs is directly coupled with the legacy system. A better foreseen approach would be to name your APIs by abstracting the backend system names as it allows seamless replacement/migration of any backend system anytime. So, this is not the correct choice too.

>> createCustomer, amendCustomer, retrieveCustomer and suspendCustomer is the right approach and is the best fit compared to other options as they are both modular and same time got the names decoupled from backend system and it has covered all requirements a System API needs.

Explanation

Correct Answer: In parallel, invoke the system API deployed to the primary environment and the system API deployed to the DR environment, and ONLY use the first response.

*****************************************

>> The API requirement in the given scenario is to respond in least possible time.

>> The option that is suggesting to first try the API in primary environment and then fallback to API in DR environment would result in successful response but NOT in least possible time. So, this is NOT a right choice of implementation for given requirement.

>> Another option that is suggesting to ONLY invoke API in primary environment and to add timeout and retries may also result in successful response upon retries but NOT in least possible time. So, this is also NOT a right choice of implementation for given requirement.

>> One more option that is suggesting to invoke API in primary environment and API in DR environment in parallel using Scatter-Gather would result in wrong API response as it would return merged results and moreover, Scatter-Gather does things in parallel which is true but still completes its scope only on finishing all routes inside it. So again, NOT a right choice of implementation for given requirement

The Correct choice is to invoke the API in primary environment and the API in DR environment parallelly, and using ONLY the first response received from one of them.

Explanation

Correct Answer: Bundle properties files for each environment into the Mule API implementation's deployable archive, then promote the Mule API implementation to the Production environment using Anypoint CLI or the Anypoint Platform REST APIs

*****************************************

>> Anypoint Exchange is for asset discovery and documentation. It has got no provision to modify the properties of Mule API implementations at all.

>> API Manager is for managing API instances, their contracts, policies and SLAs. It has also got no provision to modify the properties of API implementations.

>> API policies are to address Non-functional requirements of APIs and has again got no provision to modify the properties of API implementations.

So, the right way and recommended way to do this as part of development practice is to bundle properties files for each environment into the Mule API implementation and just point and refer to respective file per environment.

Explanation

Correct Answer: Option A

*****************************************

>> Anypoint API Manager and API policies are applicable to all types of HTTP/1.x APIs.

>> They are not applicable to WebSocket APIs, HTTP/2 APIs and gRPC APIs

Explanation

Correct Answer: A CloudHub worker fails with an out-of-memory exception.

*****************************************

>> An AWS Region itself going down will definitely result in an outage as it does not matter how many workers are assigned to the Mule App as all of those in that region will go down. This is a complete downtime and outage.

>> Extended outage of API manager during initial deployment of API implementation will of course cause issues in proper application startup itself as the API Autodiscovery might fail or API policy templates and polices may not be downloaded to embed at the time of applicaiton startup etc... there are many reasons that could cause issues.

>> A network outage onpremises would of course cause the Order Management System not accessible and it does not matter how many workers are assigned to the app they all will fail and cause outage for sure.

The only option that does NOT result in a service outage is if a cloudhub worker fails with an out-of-memory exception. Even if a worker fails and goes down, there are still other workers to handle the requests and keep the API UP and Running. So, this is the right answer.

Explanation

Correct Answer: Configure a “Worker not responding” alert in Anypoint Runtime Manager.

*****************************************

>> All the options eventually helps to generate the alert required when the application stops responding.

>> However, handling exceptions within calling API and then raising alert from API client is inappropriate and silly. There could be many API clients invoking the API implementation and it is not ideal to have this setup consistently in all of them. Not a realistic way to do.

>> Implementing a health check/ heartbeat with in the API and calling from outside to detmine the health sounds OK but needs extra setup for it and same time there are very good chances of generating false alarms when there are any intermittent network issues between external tool calling the health check API on API implementation. The API implementation itself may not have any issues but due to some other factors some false alarms may go out.

>> Creating an alert in API Manager when the API receives no requests within a specified time period would actually generate realistic alerts but even here some false alarms may go out when there are genuinely no requests from API clients.

The best and right way to achieve this requirement is to setup an alert on Runtime Manager with a condition "Worker not responding". This would generate an alert AS SOON AS the workers become unresponsive.

Bottom of Form

Top of Form

Explanation

Correct Answer: Unit tests (Whitebox)

*****************************************

Explanation

Correct Answer: Use a combination of CloudHub-deployed and manually provisioned on-premises Mule runtimes managed by the MuleSoft-hosted Platform control plane.

*****************************************

Key details to be taken from the given scenario:

>> Organization uses BOTH cloud-based and on-premises systems

>> On-premises systems can only be accessed from within the organization's intranet

Let us evaluate the given choices based on above key details:

>> CloudHub-deployed Mule runtimes can ONLY be controlled using MuleSoft-hosted control plane. We CANNOT use Private Cloud Edition's control plane to control CloudHub Mule Runtimes. So, option suggesting this is INVALID

>> Using CloudHub-deployed Mule runtimes in the shared worker cloud managed by the MuleSoft-hosted Anypoint Platform is completely IRRELEVANT to given scenario and silly choice. So, option suggesting this is INVALID

>> Using an on-premises installation of Mule runtimes that are completely isolated with NO external network access, managed by the Anypoint Platform Private Cloud Edition control plane would work for On-premises integrations. However, with NO external access, integrations cannot be done to SaaS-based apps. Moreover CloudHub-hosted apps are best-fit for integrating with SaaS-based applications. So, option suggesting this is BEST WAY.

The best way to configure and use Anypoint Platform to support these mixed/hybrid integrations is to use a combination of CloudHub-deployed and manually provisioned on-premises Mule runtimes managed by the MuleSoft-hosted Platform control plane.

Explanation

Correct Answer: At the process and system tiers

*****************************************

>> Experience Layer APIs are modeled and designed exclusively for the end user's experience. So, the data models of experience layer vary based on the nature and type of such API consumer. For example, Mobile consumers will need light-weight data models to transfer with ease on the wire, where as web-based consumers will need detailed data models to render most of the info on web pages, so on. So, enterprise data models fit for the purpose of canonical models but not of good use for experience APIs.

>> That is why, EDMs should be used extensively in process and system tiers but NOT in experience tier.

Explanation

Correct Answer: When the API implementation changes the structure of the request or response messages

*****************************************

>> RAML definition usually needs to be touched only when there are changes in the request/response schemas or in any traits on API.

>> It need not be modified for any internal changes in API implementation like performance tuning, backend system migrations etc..

Explanation

Correct Answer: From the Mule runtime irrespective of the deployment model

*****************************************

>> Monitoring and Alerting metrics are always originated from Mule Runtimes irrespective of the deployment model.

>> It may seems that some metrics (Runtime Manager) are originated from Mule Runtime and some are (API Invocations/ API Analytics) from API Manager. However, this is realistically NOT TRUE. The reason is, API manager is just a management tool for API instances but all policies upon applying on APIs eventually gets executed on Mule Runtimes only (Either Embedded or API Proxy).

>> Similarly all API Implementations also run on Mule Runtimes.

So, most of the day required for monitoring and alerts are originated fron Mule Runtimes only irrespective of whether the deployment model is MuleSoft-hosted or Customer-hosted or Hybrid.

Explanation

Correct Answer: When a new required field is added to the method called by the API client

*****************************************

>> Generally, the logic on API clients need to be updated when the API contract breaks.

>> When a new method or a child method is added to an API , the API client does not break as it can still continue to use its existing method. So these two options are out.

>> We are left for two more where "datatype of the response if changed" and "a new required field is added".

>> Changing the datatype of the response does break the API contract. However, the question is insisting on the "invocation" logic and not about the response handling logic. The API client can still invoke the API successfully and receive the response but the response will have a different datatype for some field.

>> Adding a new required field will break the API's invocation contract. When adding a new required field, the API contract breaks the RAML or API spec agreement that the API client/API consumer and API provider has between them. So this requires the API client invocation logic to also be updated.

Explanation

Correct Answer: GET, OPTIONS, HEAD

http://restcookbook.com/HTTP%20Methods/idempotency/

Explanation

Correct Answer: Set a timeout of 100ms; that leaves 400ms for other two downstream APIs to complete

*****************************************

Key details to take from the given scenario:

>> Upstream API's designed SLA is 500ms (median). Lets ignore maximum SLA response times.

>> This API calls 3 downstream APIs sequentially and all these are of similar complexity.

>> The first downstream API is offering median SLA of 100ms, 80th percentile: 500ms; 95th percentile: 1000ms.

Based on the above details:

>> We can rule out the option which is suggesting to set 50ms timeout. Because, if the median SLA itself being offered is 100ms then most of the calls are going to timeout and time gets wasted in retried them and eventually gets exhausted with all retries. Even if some retries gets successful, the remaining time wont leave enough room for 2nd and 3rd downstream APIs to respond within time.

>> The option suggesting to NOT set a timeout as the invocation of this API is mandatory and so we must wait until it responds is silly. As not setting time out would go against the good implementation pattern and moreover if the first API is not responding within its offered median SLA 100ms then most probably it would either respond in 500ms (80th percentile) or 1000ms (95th percentile). In BOTH cases, getting a successful response from 1st downstream API does NO GOOD because already by this time the Upstream API SLA of 500 ms is breached. There is no time left to call 2nd and 3rd downstream APIs.

>> It is NOT true that no timeout is possible to meet the upstream APIs desired SLA.

As 1st downstream API is offering its median SLA of 100ms, it means MOST of the time we would get the responses within that time. So, setting a timeout of 100ms would be ideal for MOST calls as it leaves enough room of 400ms for remaining 2 downstream API calls.

Explanation

Correct Answer: It makes API Manager aware of API implementations and hence enables it to enforce policies.

*****************************************

>> API Autodiscovery is a mechanism that manages an API from API Manager by pairing the deployed application to an API created on the platform.

>> API Management includes tracking, enforcing policies if you apply any, and reporting API analytics.

>> Critical to the Autodiscovery process is identifying the API by providing the API name and version.

References:

https://docs.mulesoft.com/api-manager/2.x/api-auto-discovery-new-concept

https://docs.mulesoft.com/api-manager/1.x/api-auto-discovery

https://docs.mulesoft.com/api-manager/2.x/api-auto-discovery-new-conc ept