I currently have an Azure ContainerApp architected as a BackgroundService, deployed to Azure. The Worker listens for ServiceBusMessages and processes them.
.NET6
Microsoft.ApplicationInsights.WorkerService 2.21.0
Application Insights is setup like this:
builder.Services.AddApplicationInsightsTelemetryWorkerService(opts =>
{
opts.DependencyCollectionOptions.EnableLegacyCorrelationHeadersInjection = true;
});
builder.Services.ConfigureTelemetryModule<DependencyTrackingTelemetryModule>((module, o) => { module.EnableSqlCommandTextInstrumentation = true; });
and I record the AI Operation in question by injecting a TelemetryClient and using it in the ServiceBusProcessor.ProcessMessageAsync handler like:
using (var op = _telemetryClient.StartOperation<RequestTelemetry>("ProcessAdapterMessage.ProcessSBMessage"))
{
// Process Data
}
My issue is that this operation happens a LOT, which is fine because I use AI sampling, but it's also duplicated. It's recorded once under the proper operation name "ProcessAdapterMessage.ProcessSBMessage", and once under the the operation name "<Empty>".
If I drill down into the "<Empty>" operation, it's actually just "ServiceBusProcessor.ProcessSBMessage" operations that wrap the same method as before. The reason I created the manual op is because looking at data named "Empty" isn't useful, so I'd rather keep my manual op, and make the "Empty" one go away. Any ideas on how to fix this?
This is what the "Empty" Operation details look like
This is what the "ProcessAdapterMessage.ProcessSBMessage" Operation details look like:
Related
Goal
A request comes in and is handled by the Azure Functions run-time. By default it creates a Request entry, and a bunch of Trace entries in Application Insights. I want to add a custom dimension to that top level request item (on a per-request basis) so I can use it for filtering/analysis later.
Query for -requests- on Application Insights
Resulting list of requests including custom dimensions column
The Azure Functions runtime adds a few custom dimensions already. I want to add a few of my own.
Approach
The most promising approach I've found is show below (taken from here https://github.com/microsoft/ApplicationInsights-node.js/issues/392)
appInsights.defaultClient.addTelemetryProcessor(( envelope, context ) => {
var data = envelope.data.baseData;
data.properties['mykey'] = 'myvalue';
return true;
});
However, I find that this processor is only called for requests that I initialise within my function. For example, if I make an HTTP request to another service, then details of that request will be passed thru the processor and I can add custom properties to it. But the main function does not seem to pass thru here. So I can't add my custom property.
I also tried this
defaultClient.commonProperties['anotherCustomProp'] = 'bespokeProp2'
Same problem. The custom property doesn't arrive in application insights. I've played with many variations on this and it appears that the logging done by azure-functions is walled off from anything I can do within my code.
The best workaround I have right now, is to call trackRequest manually. This is okay, except I end up with each request logged twice in application insights, one by the framework and one by me. And both need to have the same operation_id otherwise I can't find the associated trace/error items. So I'm having to extract the operationId in a slightly hacky way. This may be fine, my knowledge of application insights is pretty naive at this point.
import { setup, defaultClient } from 'applicationinsights' // i have to import the specific functions, because "import ai from applicationinsights" returns null
// call this because otherwise defaultClient is null.
// Some examples call start(), I've tried with and without this.
// I think the start() function must be useful when you're adding application-insights to a project fresh, whereas I think the azure-functions run-time must be doing this already.
setup()
const httpTrigger: AzureFunction = async function (context: Context, req: HttpRequest): Promise<void> {
// Extract the operation id from the traceparent as per w3 standard https://www.w3.org/TR/trace-context/.
const operationId = context.traceContext.traceparent.split('-')[1]
var operationIdOverride = { 'ai.operation.id': operationId }
// Create my own trackRequest entry
defaultClient.trackRequest({
name: 'my func name',
url: context.req.url.split('?')[0],
duration: 123,
resultCode: 200,
success: true,
tagOverrides: operationIdOverride,
properties: {
customProp: 'bespokeProp'
}
})
The Dream
Our C# cousins seem to have an array of options, like Activity.Current.tags and the ability to add TelemetryInitializer. However it looks like what I'm trying to do is supported, I'm just not finding the right combination of commands! Is there something similar for javascript/typescript/nodejs, where I can just add a tag on a per-request basis? Along the lines of context.traceContext.attributes['myprop'] = 'myValue'
Alternative
Alternatively, instrumenting my code using my own TelemetryClient (rather than the defaultClient) using trackRequest, trackTrace, trackError etc, is not a very big job and should work well - that would be more explicit. Should I just do that? Is there a way to disable the azure functions tracking - or perhaps I just leave that as something running side-by-side.
I know this is not recommended way by Acumatica, but we don't have other option than to use stored procedure. I have created a new processing screen to execute stored procedure but am facing time out exception.
My code sample is below:
using (new PXConnectionScope())
{
using (PXTransactionScope ts = new PXTransactionScope())
{
PXDatabase.Execute("MYSTOREDPROCEDURE", pars.ToArray());
ts.Complete();
}
}
Try executing long running code in PXLongOperation context. I assume these establishes a connection with periodic ping to avoid time-out while waiting for data to arrive.
PXLongOperation.StartOperation(Base, delegate()
{
// Code executed in long operation context
});
If your code is executed from the context of a processing delegate I think it should be already wrapped in a long operation though. Otherwise long operation should be used inside an action event handler.
A last recourse would be to increase time-out in the web.config file.
Use of stored procedure is a concern mainly for SAAS hosting and obtaining an Acumatica ISV Certification. There's likely no official support for it but I doubt it's gonna go away.
My Node app uses Mongo change streams, and the app runs 3+ instances in production (more eventually, so this will become more of an issue as it grows). So, when a change comes in the change stream functionality runs as many times as there are processes.
How to set things up so that the change stream only runs once?
Here's what I've got:
const options = { fullDocument: "updateLookup" };
const filter = [
{
$match: {
$and: [
{ "updateDescription.updatedFields.sites": { $exists: true } },
{ operationType: "update" }
]
}
}
];
const sitesStream = Client.watch(sitesFilter, options);
// Start listening to site stream
sitesStream.on("change", async change => {
console.log("in site change stream", change);
console.log(
"in site change stream, update desc",
change.updateDescription
);
// Do work...
console.log("site change stream done.");
return;
});
It can easily be done with only Mongodb query operators. You can add a modulo query on the ID field where the divisor is the number of your app instances (N). The remainder is then an element of {0, 1, 2, ..., N-1}. If your app instances are numbered in ascending order from zero to N-1 you can write the filter like this:
const filter = [
{
"$match": {
"$and": [
// Other filters
{ "_id": { "$mod": [<number of instances>, <this instance's id>]}}
]
}
}
];
Doing this with strong guarantees is difficult but not impossible. I wrote about the details of one solution here: https://www.alechenninger.com/2020/05/building-kafka-like-message-queue-with.html
The examples are in Java but the important part is the algorithm.
It comes down to a few techniques:
Each process attempts to obtain a lock
Each lock (or each change) has an associated fencing token
Processing each change must be idempotent
While processing the change, the token is used to ensure ordered, effectively-once updates.
More details in the blog post.
It sounds like you need a way to partition updates between instances. Have you looked into Apache Kafka? Basically what you would do is have a single application that writes the change data to a partitioned Kafka Topic and have your node application be a Kafka consumer. This would ensure only one application instance ever receives an update.
Depending on your partitioning strategy, you could even ensure that updates for the same record always go to the same node app (if your application needs to maintain its own state). Otherwise, you can spread out the updates in a round robin fashion.
The biggest benefit to using Kafka is that you can add and remove instances without having to adjust configurations. For example, you could start one instance and it would handle all updates. Then, as soon as you start another instance, they each start handling half of the load. You can continue this pattern for as many instances as there are partitions (and you can configure the topic to have 1000s of partitions if you want), that is the power of the Kafka consumer group. Scaling down works in the reverse.
While the Kafka option sounded interesting, it was a lot of infrastructure work on a platform I'm not familiar with, so I decided to go with something a little closer to home for me, sending an MQTT message to a little stand alone app, and letting the MQTT server monitor messages for uniqueness.
siteStream.on("change", async change => {
console.log("in site change stream);
const mqttClient = mqtt.connect("mqtt://localhost:1883");
const id = JSON.stringify(change._id._data);
// You'll want to push more than just the change stream id obviously...
mqttClient.on("connect", function() {
mqttClient.publish("myTopic", id);
mqttClient.end();
});
});
I'm still working out the final version of the MQTT server, but the method to evaluate uniqueness of messages will probably store an array of change stream IDs in application memory, as there is no need to persist them, and evaluate whether to proceed any further based on whether that change stream ID has been seen before.
var mqtt = require("mqtt");
var client = mqtt.connect("mqtt://localhost:1883");
var seen = [];
client.on("connect", function() {
client.subscribe("myTopic");
});
client.on("message", function(topic, message) {
context = message.toString().replace(/"/g, "");
if (seen.indexOf(context) < 0) {
seen.push(context);
// Do stuff
}
});
This doesn't include security, etc., but you get the idea.
Will that having a field in DB called status which will be updated using findAnUpdate based on the event received from change stream. So lets say you get 2 events at the same time from change stream. First event will update the status to start and the other will throw error if status is start. So the second event will not process any business logic.
I'm not claiming those are rock-solid production grade solutions, but I believe something like this could work
Solution 1
applying Read-Modify-Write:
Add version field to the document, all the created docs have version=0
Receive ChangeStream event
Read the document that needs to be updated
Perform the update on the model
Increment version
Update the document where both id and version match, otherwise discard the change
Yes, it creates 2 * n_application_replicas useless queries, so there is another option
Solution 2
Create collection of ResumeTokens in mongo which would store collection -> token mapping
In the changeStream handler code, after successful write, update ResumeToken in the collection
Create a feature toggle that will disable reading ChangeStream in your application
Configure only a single instance of your application to be a "reader"
In case of "reader" failure you might either enable reading on another node, or redeploy the "reader" node.
As a result: there might be an infinite amount of non-reader replicas and there won't be any useless queries
We've been developing using Excel JavaScript API for quite a few months now. We have been coming across context related issues which got resolved for unknown reasons. We weren't able to replicate these issues and wondered how they got resolved. Recently similar issues have started popping up again.
Error we consistently get:
property 'address' is not available. Before reading the property's
value, call the load method on the containing object and call
"context.sync()" on the associated request context.
We thought as we have multiple functions defined to modularise code in project, may be context differs somewhere among these functions which has gone unnoticed. So we came up with single context solution implemented via JavaScript Module pattern.
var ContextManager = (function () {
var xlContext;//single context for entire project/application.
function loadContext() {
xlContext = new Excel.RequestContext();
}
function sync(object) {
return (object === undefined) ? xlContext.sync() : xlContext.sync(object);
}
function getWorksheetByName(name) {
return xlContext.workbook.worksheets.getItem(name.toString());
}
//public
return {
loadContext: loadContext,
sync: sync,
getWorksheetByName: getWorksheetByName
};
})();
NOTE: above code shortened. There are other methods added to ensure that single context gets used throughout application.
While implementing single context, this time round, we have been able to replicate the issue though.
Office.initialize = function (reason) {
$(document).ready(function () {
ContextManager.loadContext();
function loadRangeAddress(rng, index) {
rng.load("address");
ContextManager.sync().then(function () {
console.log("Address: " + rng.address);
}).catch(function (e) {
console.log("Failed address for index: " + index);
});
}
for (var i = 1; i <= 1000; i++) {
var sheet = ContextManager.getWorksheetByName("Sheet1");
loadRangeAddress(sheet.getRange("A" + i), i);//I expect to see a1 to a1000 addresses in console. Order doesn't matter.
}
});
}
In above case, only "A1" gets printed as range address to console. I can't see any of the other addresses (A2 to A1000)being printed. Only catch block executes. Can anyone explain why this happens?
Although I've written for loop above, that isn't my use case. In real use case, such situations occur where one range object in function a needs to load range address. In the mean while another function b also wants to load range address. Both function a and function b work asynchronously on separate tasks such as one creates table object (table needs address) and other pastes data to sheet (there's debug statement to see where data was pasted).
This is something our team hasn't been able to figure out or find a solution for.
There is a lot packed into this code, but the issue you have is that you're calling sync a whole bunch of times without awaiting the previous sync.
There are several problems with this:
If you were using different contexts, you would actually see that there is a limit of ~50 simultaneous requests, after which you'll get errors.
In your case, you're running into a different (and almost opposite) problem. Given the async nature of the APIs, and the fact that you're not awaiting on the sync-s, your first sync request (which you'd think is for just A1) will actually contain all the load requests from the execution of the entire for loop. Now, once this first sync is dispatched, the action queue will be cleared. Which means that your second, third, etc. sync will see that there is no pending work, and will no-op, executing before the first sync ever came back with the values!
[This might be considered a bug, and I'll discuss with the team about fixing it. But it's still a very dangerous thing to not await the completion of a sync before moving on to the next batch of instructions that use the same context.]
The fix is to await the sync. This is far and away the simplest to do in TypeScript 2.1 and its async/await feature, otherwise you need to do the async version of the for loop, which you can look up, but it's rather unintuitive (requires creating an uber-promise that keeps chaining a bunch of .then-s)
So, your modified TypeScript-ified code would be
ContextManager.loadContext();
async function loadRangeAddress(rng, index) {
rng.load("address");
await ContextManager.sync().then(function () {
console.log("Address: " + rng.address);
}).catch(function (e) {
OfficeHelpers.Utilities.log(e);
});
}
for (var i = 1; i <= 1000; i++) {
var sheet = ContextManager.getWorksheetByName("Sheet1");
await loadRangeAddress(sheet.getRange("A" + i), i);//I expect to see a1 to a1000 addresses in console. Order doesn't matter.
}
Note the async in front of the loadRangeAddress function, and the two await-s in front of ContextManager.sync() and loadRangeAddress.
Note that this code will also run quite slowly, as you're making an async roundtrip for each cell. Which means you're not using batching, which is at the very core of the object-model for the new APIs.
For completeness sake, I should also note that creating a "raw" RequestContext instead of using Excel.run has some disadvantages. Excel.run does a number of useful things, the most important of which is automatic object tracking and un-tracking (not relevant here, since you're only reading back data; but would be relevant if you were loading and then wanting to write back into the object).
Finally, if I may recommend (full disclosure: I am the author of the book), you will probably find a good bit of useful info about Office.js in the e-book "Building Office Add-ins using Office.js", available at https://leanpub.com/buildingofficeaddins. In particular, it has a very detailed (10-page) section on the internal workings of the object model ("Section 5.5: Implementation details, for those who want to know how it really works"). It also offers advice on using TypeScript, has a general Promise/async-await primer, describes what .run does, and has a bunch more info about the OM. Also, though not available yet, it will soon offer information on how to resume using the same context (using a newer technique than what was originally described in How can a range be used across different Word.run contexts?). The book is a lean-published "evergreen" book, son once I write the topic in the coming weeks, an update will be available to all existing readers.
Hope this helps!
The Azure Service Bus supports a built-in retry mechanism which makes an abandoned message immediately visible for another read attempt. I'm trying to use this mechanism to handle some transient errors, but the message is made available immediately after being abandoned.
What I would like to do is make the message invisible for a period of time after it is abandoned, preferably based on an exponentially incrementing policy.
I've tried to set the ScheduledEnqueueTimeUtc property when abandoning the message, but it doesn't seem to have an effect:
var messagingFactory = MessagingFactory.CreateFromConnectionString(...);
var receiver = messagingFactory.CreateMessageReceiver("test-queue");
receiver.OnMessageAsync(async brokeredMessage =>
{
await brokeredMessage.AbandonAsync(
new Dictionary<string, object>
{
{ "ScheduledEnqueueTimeUtc", DateTime.UtcNow.AddSeconds(30) }
});
}
});
I've considered not abandoning the message at all and just letting the lock expire, but this would require having some way to influence how the MessageReceiver specifies the lock duration on a message, and I can't find anything in the API to let me change this value. In addition, it wouldn't be possible to read the delivery count of the message (and therefore make a decision for how long to wait for the next retry) until after the lock is already required.
Can the retry policy in the Message Bus be influenced in some way, or can a delay be artificially introduced in some other way?
Careful here because I think you are confusing the retry feature with the automatic Complete/Abandon mechanism for the OnMessage event-driven message handling. The built in retry mechanism comes into play when a call to the Service Bus fails. For example, if you call to set a message as complete and that fails, then the retry mechanism would kick in. If you are processing a message an exception occurs in your own code that will NOT trigger a retry through the retry feature. Your question doesn't get explicit on if the error is from your code or when attempting to contact the service bus.
If you are indeed after modifying the retry policy that occurs when an error occurs attempting to communicate with the service bus you can modify the RetryPolicy that is set on the MessageReciver itself. There is an RetryExponitial which is used by default, as well as an abstract RetryPolicy you can create your own from.
What I think you are after is more control over what happens when you get an exception doing your processing, and you want to push off working on that message. There are a few options:
When you create your message handler you can set up OnMessageOptions. One of the properties is "AutoComplete". By default this is set to true, which means as soon as processing for the message is completed the Complete method is called automatically. If an exception occurs then abandon is automatically called, which is what you are seeing. By setting the AutoComplete to false you required to call Complete on your own from within the message handler. Failing to do so will cause the message lock to eventually run out, which is one of the behaviors you are looking for.
So, you could write your handler so that if an exception occurs during your processing you simply do not call Complete. The message would then remain on the queue until it's lock runs out and then would become available again. The standard dead lettering mechanism applies and after x number of tries it will be put into the deadletter queue automatically.
A caution of handling this way is that any type of exception will be treated this way. You really need to think about what types of exceptions are doing this and if you really want to push off processing or not. For example, if you are calling a third party system during your processing and it gives you an exception you know is transient, great. If, however, it gives you an error that you know will be a big problem then you may decide to do something else in the system besides just bailing on the message.
You could also look at the "Defer" method. This method actually will then not allow that message to be processed off the queue unless it is specifically pulled by its sequence number. You're code would have to remember the sequence number value and pull it. This isn't quite what you described though.
Another option is you can move away from the OnMessage, Event-driven style of processing messages. While this is very helpful you don't get a lot of control over things. Instead hook up your own processing loop and handle the abandon/complete on your own. You'll also need to deal some of the threading/concurrent call management that the OnMessage pattern gives you. This can be more work but you have the ultimate in flexibility.
Finally, I believe the reason the call you made to AbandonAsync passing the properties you wanted to modify didn't work is that those properties are referring to Metadata properties on the method, not standard properties on BrokeredMessage.
I actually asked this same question last year (implementation aside) with the three approaches I could think of looking at the API. #ClemensVasters, who works on the SB team, responded that using Defer with some kind of re-receive is really the only way to control this precisely.
You can read my comment to his answer for a specific approach to doing it where I suggest using a secondary queue to store messages that indicate which primary messages have been deferred and need to be re-received from the main queue. Then you can control how long you wait by setting the ScheduledEnqueueTimeUtc on those secondary messages to control exactly how long you wait before you retry.
I ran into a similar issue where our order picking system is legacy and goes into maintenance mode each night.
Using the ideas in this article(https://markheath.net/post/defer-processing-azure-service-bus-message) I created a custom property to track how many times a message has been resubmitted and manually dead lettering the message after 10 tries. If the message is under 10 retries it clones the message increments the custom property and sets the en queue of the new message.
using Microsoft.Azure.ServiceBus;
public PickQueue()
{
queueClient = new QueueClient(QUEUE_CONN_STRING, QUEUE_NAME);
}
public async Task QueueMessageAsync(int OrderId)
{
string body = JsonConvert.SerializeObject(OrderId);
var message = new Message(Encoding.UTF8.GetBytes(body));
await queueClient.SendAsync(message);
}
public async Task ReQueueMessageAsync(Message message, DateTime utcEnqueueTime)
{
int resubmitCount = (int)(message.UserProperties["ResubmitCount"] ?? 0) + 1;
if (resubmitCount > 10)
{
await queueClient.DeadLetterAsync(message.SystemProperties.LockToken);
}
else
{
Message clone = message.Clone();
clone.UserProperties["ResubmitCount"] = ++resubmitCount;
await queueClient.ScheduleMessageAsync(message, utcEnqueueTime);
}
}
This question asks how to implement exponential backoff in Azure Functions. If you do not want to use the built-in RetryPolicy (only available when autoComplete = false), here's the solution I've been using:
public static async Task ExceptionHandler(IMessageSession MessageSession, string LockToken, int DeliveryCount)
{
if (DeliveryCount < Globals.MaxDeliveryCount)
{
var DelaySeconds = Math.Pow(Globals.ExponentialBackoff, DeliveryCount);
await Task.Delay(TimeSpan.FromSeconds(DelaySeconds));
await MessageSession.AbandonAsync(LockToken);
}
else
{
await MessageSession.DeadLetterAsync(LockToken);
}
}