protobuf-net concurrent performance issue in TakeLock - protocol-buffers

We're using protobuf-net for sending log messages between services. When profiling stress testing, under high concurrency, we see very high CPU usage and that TakeLock in RuntimeTypeModel is the culprit. The hot call stack looks something like:
*Our code...*
ProtoBuf.Serializer.SerializeWithLengthPrefix(class System.IO.Stream,!!0,valuetype ProtoBuf.PrefixStyle)
ProtoBuf.Serializer.SerializeWithLengthPrefix(class System.IO.Stream,!!0,valuetype ProtoBuf.PrefixStyle,int32)
ProtoBuf.Meta.TypeModel.SerializeWithLengthPrefix(class System.IO.Stream,object,class System.Type,valuetype ProtoBuf.PrefixStyle,int32)
ProtoBuf.Meta.TypeModel.SerializeWithLengthPrefix(class System.IO.Stream,object,class System.Type,valuetype ProtoBuf.PrefixStyle,int32,class ProtoBuf.SerializationContext)
ProtoBuf.ProtoWriter.WriteObject(object,int32,class ProtoBuf.ProtoWriter,valuetype ProtoBuf.PrefixStyle,int32)
ProtoBuf.BclHelpers.WriteNetObject(object,class ProtoBuf.ProtoWriter,int32,valuetype
ProtoBuf.BclHelpers/NetObjectOptions)
ProtoBuf.Meta.TypeModel.GetKey(class System.Type&)
ProtoBuf.Meta.RuntimeTypeModel.GetKey(class System.Type,bool,bool)
ProtoBuf.Meta.RuntimeTypeModel.FindOrAddAuto(class System.Type,bool,bool,bool)
ProtoBuf.Meta.RuntimeTypeModel.TakeLock(int32&)
[clr.dll]
I see that we can use the new precompiler to get a speed boost, but I'm wondering if that will get rid of the issue (sounds like it doesn't use reflection); it would be a bit of work for me to integrate this, so I haven't tested it yet. I also see the option to call Serializer.PrepareSerializer. My initial (small scale) testing didn't make the prepare seem promising.
A little more info about the type we're serializing:
[ProtoContract]
public class SomeMessage
{
[ProtoMember(1)]
public SomeEnumType SomeEnum { get; set; }
[ProtoMember(2)]
public long SomeId{ get; set; }
[ProtoMember(3)]
public string SomeString{ get; set; }
[ProtoMember(4)]
public DateTime SomeDate { get; set; }
[ProtoMember(5, DynamicType = true, OverwriteList = true)]
public Collection<object> SomeArguments
}
Thanks for your help!
UPDATE 9/17
Thanks for your response! We're going to try the workaround you suggest and see if that fixes things.
This code lives in our logging system so, in the SomeMessage example, SomeString is really a format string (e.g. "Hello {0}") and the SomeArguments collection is a list of objects used to fill in the format string, just like String.Format. Before we serialize, we look at each argument and call DynamicSerializer.IsKnownType(argument.GetType()), if it isn't known, we convert it to a string first. I haven't looked at the ratios of data, but I'm pretty sure we have a lot of different strings coming in as arguments.
Let me know if this helps. If you need, I'll try to get more details.

TakeLock is only used when it is changing the model, for example because it is seeing a type for the first time. You shouldn't normally see TakeLock after the first time a particular type has been used. In most cases, using Serializaer.PrepareSerializer<SomeMessage>() should perform all the necessary initialization (and similar for any other contracts you are using).
However! I wonder if perhaps this is also related to your use of DynamicType; what are the actual objects being used here? It might be that I need to tweak the logic here, so that it doesn't spend any time on that step. If you let me know the actual objects (so I can repro), I will try to run some tests.
As for whether the precompiler would change this; yes it would. A fully compiled static model has a completely different implementation of the ProtoBuf.Meta.TypeModel.GetKey method, so it would never call TakeLock (you don't need to protect a model that can never change!). But you can actuallydo something very similar without needing to use precompile. Consider the following, run as part of your app's initialization:
static readonly TypeModel serializer;
...
var model = TypeModel.Create();
model.Add(typeof(SomeMessage), true);
// TODO add other contracts you use here
serializer = model.Compile();
This will create a fully static-compiled serializer assembly in memory (instead of a mutable model with individual operations compiled). If you now use serializer.Serialize(...) instead of Serializer.Serialize (i.e. the instance method on your stored TypeModel rather than the static method on Serializer) then it will essentially be doing something very similar to "precompiler", but without the need to actualy precompile it (obviously this will only be available on "full" .NET). This will then never call TakeLock, as it is running a fixed model, rather than a flexible model. It does, however, require you to know what contract-types you use. You could use reflection to find these, by looking for all those types with a given attribute:
static readonly TypeModel serializer;
...
var model = TypeModel.Create();
Type attributeType = typeof(ProtoContractAttribute);
foreach (var type in typeof(SomeMessage).Assembly.GetTypes()) {
if (Attribute.IsDefined(type, attributeType)) {
model.Add(type, true);
}
}
serializer = model.Compile();
But emphasis: the above is a workaround; it sounds like there's a glitch, which I'll happily investigate if I can see an example where it actually happens; most importantly: what are the objects in SomeArguments?

Related

Why do static members lose their value in Xamarin.Forms

I’m having issues with a static member of my app class losing its value and I’m not quite sure I understand why. In my app constructor I check if the user is logged in and if not redirect to a login page where I set the static app class member.
I understand if the app is forced to close to free up resources, these values are not retained so a new app instance would start and go back to login screen. However, what I’m seeing is the static member losing its value during an application session. I can do a check to see if this is null on resume and redirect to login page but I don’t understand why this happens.
My understaning was that the only way you would lose values would be if the app was killed in the background but this problem would suggest it can happen when resuming too.
In a normal C# application static members will typically survive forever, but unfortunately your observations are entirely correct; in Xamarin Forms static members are not guaranteed to persist for the length of the application's life.
In Android's case if the underlying platform indicates a low memory state (or increased demands on memory from multiple running applications) then static members are considered collectable by the GC, which is often triggered when you pause the application (ie. switching to a different app). They will be reduced to their default value, eg. null, zero, etc.
I've wrestled with this curio for years, and the most performant work around is to implement a re-population pattern on those static members, eg.
internal List<MyCustomType> _AListOfStuff
internal List<MyCustomType> AListOfStuff
{
get
{
if (_AListOfStuff == null)
{
PopulateAListOfStuff(); //If this occurs then the static member has been garbage collected: reload it
}
return _AListOfStuff;
}
}
From what you've said, I appreciate that your particular usage of static members probably doesn't fit with this solution, however all I can offer is that you're not crazy; it is a documented quirk, and not considered a bug (don't even bother shaking that tree; I've been down that route with the devs and was told in no uncertain terms that the behaviour is here to stay, and is necessary to ensure overall device stability).
Static member will not lose. If we see the code then we can assist further. Another approach would be, try using singleton pattern, it will create new instance only if it's instance is null. sample below:
public sealed class SingletonSample
{
private static SingletonSample instance = null;
private static readonly object padlock = new object();
public static SingletonSample Instance
{
get
{
lock (padlock)
{
if (instance == null)
{
instance = new SingletonSample();
}
return instance;
}
}
}
public string FirstName { get; set; }
}

MongoDB - override default Serializer for a C# primitive type

I'd like to change the representation of C# Doubles to rounded Int64 with a four decimal place shift in the serialization C# Driver's stack for MongoDB. In other words, store (Double)29.99 as (Int64)299900
I'd like this to be transparent to my app. I've had a look at custom serializers but I don't want to override everything and then switch on the Type with fallback to the default, as that's a bit messy.
I can see that RegisterSerializer() won't let me add one for an existing type, and that BsonDefaultSerializationProvider has a static list of primitive serializers and it's marked as internal with private members so I can't easily subclass.
I can also see that it's possible to RepresentAs Int64 for Doubles, but this is a cast not a conversion. I need essentially a cast AND a conversion in both serialization directions.
I wish I could just give the default serializer a custom serializer to override one of it's own, but that would mean a dirty hack.
Am I missing a really easy way?
You can definitely do this, you just have to get the timing right. When the driver starts up there are no serializers registered. When it needs a serializer, it looks it up in the dictionary where it keeps track of the serializers it knows about (i.e. the ones that have been registered). Only it it can't find one in the dictionary does it start figuring out where to get one (including calling the serialization providers) and if it finds one it registers it.
The limitation in RegisterSerializer is there so that you can't replace an existing serializer that has already been used. But that doesn't mean you can't register your own if you do it early enough.
However, keep in mind that registering a serializer is a global operation, so if you register a custom serializer for double it will be used for all doubles, which could lead to unexpected results!
Anyway, you could write the custom serializer something like this:
public class CustomDoubleSerializer : BsonBaseSerializer
{
public override object Deserialize(BsonReader bsonReader, Type nominalType, Type actualType, IBsonSerializationOptions options)
{
var rep = bsonReader.ReadInt64();
return rep / 100.0;
}
public override void Serialize(BsonWriter bsonWriter, Type nominalType, object value, IBsonSerializationOptions options)
{
var rep = (long)((double)value * 100);
bsonWriter.WriteInt64(rep);
}
}
And register it like this:
BsonSerializer.RegisterSerializer(typeof(double), new CustomDoubleSerializer());
You could test it using the following class:
public class C
{
public int Id;
public double X;
}
and this code:
BsonSerializer.RegisterSerializer(typeof(double), new CustomDoubleSerializer());
var c = new C { Id = 1, X = 29.99 };
var json = c.ToJson();
Console.WriteLine(json);
var r = BsonSerializer.Deserialize<C>(json);
Console.WriteLine(r.X);
You can also use your own serialization provider to tell Mongo which serializer to use for certain types, which I ended up doing to mitigate some of the timing issues mentioned when trying to override existing serializers. Here's an example of a serialisation provider that overrides how to serialize decimals:
public class CustomSerializationProvider : IBsonSerializationProvider
{
public IBsonSerializer GetSerializer(Type type)
{
if (type == typeof(decimal)) return new DecimalSerializer(BsonType.Decimal128);
return null; // falls back to Mongo defaults
}
}
If you return null from your custom serialization provider, it will fall back to using Mongo's default serialization provider.
Once you've written your provider, you just need to register it:
BsonSerializer.RegisterSerializationProvider(new CustomSerializationProvider());
I looked through the latest iteration of the driver's code and checked if there's some sort of backdoor to set custom serializers. I am afraid there's none; you should open an issue in the project's bug tracker if you think this needs to be looked at for future iterations of the driver (https://jira.mongodb.org/).
Personally, I'd open a ticket -- and if a quick workaround is necessary or required, I'd subclass DoubleSerializer, implement the new behavior, and then use Reflection to inject it into either MongoDB.Bson.Serialization.Serializers.DoubleSerializer.__instance or MongoDB.Bson.Serialization.BsonDefaultSerializationProvider.__serializers.

Can someone help me understand Guava CacheLoader?

I'm new to Google's Guava library and am interested in Guava's Caching package. Currently I have version 10.0.1 downloaded. After reviewing the documentation, the JUnit tests source code and even after searching google extensively, I still can't figure out how to use the Caching package. The documentation is very short, as if it was written for someone who has been using Guava's library not for a newbie like me. I just wish there are more real world examples on how to use Caching package propertly.
Let say I want to build a cache of 10 non expiring items with Least Recently Used (LRU) eviction method. So from the example found in the api, I build my code like the following:
Cache<String, String> mycache = CacheBuilder.newBuilder()
.maximumSize(10)
.build(
new CacheLoader<String, String>() {
public String load(String key) throws Exception {
return something; // ?????
}
});
Since the CacheLoader is required, I have to include it in the build method of CacheBuilder. But I don't know how to return the proper value from mycache.
To add item to mycache, I use the following code:
mycache.asMap().put("key123", "value123");
To get item from mycache, I use this method:
mycache.get("key123")
The get method will always return whatever value I returned from CacheLoader's load method instead of getting the value from mycache. Could someone kindly tell me what I missed?
Guava's Cache type is generally intended to be used as a computing cache. You don't usually add values to it manually. Rather, you tell it how to load the expensive to calculate value for a key by giving it a CacheLoader that contains the necessary code.
A typical example is loading a value from a database or doing an expensive calculation.
private final FooDatabase fooDatabase = ...;
private final LoadingCache<Long, Foo> cache = CacheBuilder.newBuilder()
.maximumSize(10)
.build(new CacheLoader<Long, Foo>() {
public Foo load(Long id) {
return fooDatabase.getFoo(id);
}
});
public Foo getFoo(long id) {
// never need to manually put a Foo in... will be loaded from DB if needed
return cache.getUnchecked(id);
}
Also, I tried the example you gave and mycache.get("key123") returned "value123" as expected.

How to take advantage of an auto-property when refactoring this .Net 1.1 sample?

I see a lot of legacy .Net 1.1-style code at work like in example below, which I would like to shrink with the help of an auto-property. This will help many classes shrink by 30-40%, which I think would be good.
public int MyIntThingy
{
get
{
return _myIntThingy;
}
set
{
_myIntThingy = value;
}
} private int _myIntThingy = -1;
This would become:
public int MyIntThingy
{
get;
set;
}
And the only question is - where do I set MyIntThingy = -1;?
If I wrote the class from the start, then I would have a better idea, but I did not. An obvious answer would be: put it in the constructor. Trouble is: there are many constructors in this class. Watching the initialization to -1 in the debugger, I see it happen (I believe) before the constructor gets called. It is almost as if I need to use a static constructor as described here:
http://www.c-sharpcorner.com/uploadfile/cupadhyay/staticconstructors11092005061428am/staticconstructors.aspx
except that my variables are not static. Java's static initializer comes to mind, but again - my variables are not static. http://www.glenmccl.com/tip_003.htm
I want to make stylistic but not functional changes to this class. As crappy as it is, it has been tested and working for a few years now. breaking the functionality would be bad. So ... I am looking for shorter, sweeter, cuter, and yet EQUIVALENT code. Let me know if you have questions.
I'm afraid that you have no option.
If you want to use an auto-property with an initial value that differs from the type's default value then you'll need to set the initial value in the constructor(s).
If you just need a stylistic, non-breaking change, consider changing the format a little:
public int MyIntThingy
{
get { return _myIntThingy; }
set { _myIntThingy = value; }
}
private int _myIntThingy = -1;
Isn't that prettier?
And consider using auto-properties for future code only. It sounds too risky to use them on existing code, unless there are no default values.

Enterprise Library Validation Block - Should validation be placed on class or interface?

I am not sure where the best place to put validation (using the Enterprise Library Validation Block) is? Should it be on the class or on the interface?
Things that may effect it
Validation rules would not be changed in classes which inherit from the interface.
Validation rules would not be changed in classes which inherit from the class.
Inheritance will occur from the class in most cases - I suspect some fringe cases to inherit from the interface (but I would try and avoid it).
The interface main use is for DI which will be done with the Unity block.
The way you are trying to use the Validation Block with DI, I dont think its a problem if you set the attributes at interface level. Also, I dont think it should create problems in the inheritance chain. However, I have mostly seen this block used at class level, with an intent to keep interfaces not over specify things. IMO i dont see a big threat in doing this.
Be very careful here, your test is too simple.
This will not work as you expect for SelfValidation Validators or Class Validators, only for the simple property validators like you have there.
Also, if you are using the PropertyProxyValidator in an ASP.NET page, iI don;t believe it will work either, because it only looks a field validators, not inherited/implemented validators...
Yes big holes in the VAB if you ask me..
For the sake of completeness I decided to write a small test to make sure it would work as expected and it does, I'm just posting it here in case anyone else wants it in future.
using System;
using Microsoft.Practices.EnterpriseLibrary.Validation;
using Microsoft.Practices.EnterpriseLibrary.Validation.Validators;
namespace ConsoleApplication1
{
class Program
{
static void Main(string[] args)
{
ISpike spike = new Spike();
spike.Name = "A really long name that will fail.";
ValidationResults r = Validation.Validate<ISpike>(spike);
if (!r.IsValid)
{
throw new InvalidOperationException("Validation error found.");
}
}
}
public class Spike : ConsoleApplication1.ISpike
{
public string Name { get; set; }
}
interface ISpike
{
[StringLengthValidator(2, 5)]
string Name { get; set; }
}
}
What version of Enterprise Library are you using for your code example? I tried it using Enterprise Library 5.0, but it didn't work.
I tracked it down to the following section of code w/in the EL5.0 source code:
[namespace Microsoft.Practices.EnterpriseLibrary.Validation]
[public static class Validation]
public static ValidationResults Validate<T>(T target, ValidationSpecificationSource source)
{
Type targetType = target != null ? target.GetType() : typeof(T);
Validator validator = ValidationFactory.CreateValidator(targetType, source);
return validator.Validate(target);
}
If the target object is defined, then target.GetType() will return the most specific class definition, NOT the interface definition.
My workaround is to replace your line:
ValidationResults r = Validation.Validate<ISpike>(spike);
With:
ValidationResults r ValidationFactory.CreateValidator<ISpike>().Validate(spike);
This got it working for me.

Resources