perf_event_open: handling last recorded sample - linux-kernel

When counting for events based on a specific sampling period, how to handle the last recorded sample when the last counter value of the leader is less than the sampling period.
Update:
I have checked the value of type which is a member of struct perf_event_header. For the last recorded sample this value is zero and according to perf_event.h header file, it does not seem that the value of zero has a corresponding sample record type!
To put my question in other words: How does perf_event API deal with the case when the workload finishes execution but the group leader counter value is less than the value of the sampling period? Is the data discarded at this case?

How does perf_event API deal with the case when the workload finishes execution but the group leader counter value is less than the value of the sampling period?
Nothing happens. If the event count is not reached yet, no sample is written.
You should consider that samples are typically statistical information.
If you really need to know you could possibly use some form of ptrace and manually read the counter value before the thread terminates.
If you read a perf_event_header with a type == 0, I would be concerned. I don't think that should ever happen.
Edit:
As per the manpage, I believe you cannot read the remaining value from that particular event because sampling and counting events are exclusive.
Events come in two flavors: counting and sampled. A counting event
one that is used for counting the aggregate number of events that.
In general, counting event results are gathered with a
read(2) call. A sampling event periodically writes measurements to a buffer
that can then be accessed via mmap(2).

Related

Sequential data output without writing to disk

4kb memory
1gb data
8 bytes per data
Sequential data output without writing to disk.
First, you can only accumulate and output a maximum of 4kb per time through the data. So you will need at least something like 250,000 passes.
How can you accumulate stuff in a pass?
In pseudocode the idea is like this.
while not done:
for each data (8 bytes) in dataset:
if this data has never been output:
if it might belong in the current batch:
add to current batch (evicting something else if needed)
if current_batch not empty:
sort current batch
emit current batch
update "never been output" filter
else:
done
What does that filter look like? It needs to know three things:
What is the maximum value so far emitted?
How many times has it been emitted?
How many times has it been seen on this pass?
Any value below the maximum value gets ignored. After you've seen the value enough times, you can add it to the current batch.
Now how about the current batch you're accumulating? That can be a heap that tells you the maximum value in the batch. If the heap is not full, or if the current value is below the maximum in the batch, you add it to the batch and lose the current max.
If the heap is arranged in memory so that the smallest is first, when the batch is done you can remove the max, which will free up the last slot (that's how heaps work), and put the max there. Keep doing that and you'll heapsort the batch. Now you can easily update the filter, and then emit the batch.
I don't think you can get significantly more efficient than this.
If I was asked this in an interview, I'd know the answer, but I'd also see being asked the question as a sign that the company's hiring process is suboptimal. This would make me less inclined to be hired there unless there was some purpose I could see to why they hired this way. (I know why FAANGs do. But at most companies I'd call it a red flag.)

Approach to measuring end-to-end latency from a sales transaction to a stock level in a database

I have a system in which sales transactions are written to a Kafka topic in real time. One of the consumers of this data is an aggregator program which maintains a database of stock quantities for all locations, in real time; it will consume data from multiple other sources as well. For example, when a product is sold from a store, the aggregator will reduce the quantity of that product in that store by the quantity sold.
This aggregator's database will be presented via an API to allow applications to check stock availability (the inventory) in any store in real time.
(Note for context - yes, there is an ERP behind all this which does a lot more; the purpose of this inventory API is to consume data from multiple sources, including the ERP and the ERP's data feeds, and potentially other ERPs in future, to give a single global information source for this singular purpose).
What I want to do is to measure the end-to-end latency: how long it takes from a sales transaction being written to the topic, to being processed by the aggregator (not just read from the topic). This will give an indicator of how far behind real-time the inventory database is.
The sales transaction topic will probably be partitioned, so the transactions may not arrive in order.
So far I have thought of two methods.
Method 1 - measure latency via stock level changes
Here, the sales producer injects a special "measurement" sale each minute, for an invalid location like "SKU 0 in branch 0". The sale quantity would be based on the time of day, using a numerical sequence of some kind. A program would then poll the inventory API, or directly read the database, to check for changes in the level. When it changes, the magnitude of the change will indicate the time of the originating transaction. The difference between then and now gives us the latency.
Problem: If multiple transactions are queued and are then later all processed together, the change in inventory value will be the sum of the queued transactions, giving a false reading.
To solve this, the sequence of numbers would have to be chosen such that when they are added together, we can always determine which was the lowest number, giving us the oldest transaction and therefore the latency.
We could use powers of 2 for this, so the lowest bit set would indicate the earliest transaction time. Our sequence would have to reset every 30 or 60 minutes and we'd have to cope with wraparound and lots of edge cases.
Assuming we can solve the wraparound problem and that a maximum measurable latency of, say, 20 minutes is OK (after which we just say it's "too high"), then with this method, it does not matter whether transactions are processed out of sequence or split into partitions.
This method gives a "true" picture of the end-to-end latency, in that it's measuring the point at which the database has actually been updated.
Method 2 - measure latency via special timestamp record
Instead of injecting measurement sales records, we use a timestamp which the producer is adding to the raw data. This timestamp is just the time at which the producer transmitted this record.
The aggregator would maintain a measurement of the most recently seen timestamp. The difference between that and the current time would give the latency.
Problem: If transactions are not processed in order, the latency measurement will be unstable, because it relies on the timestamps arriving in sequence.
To solve this, the aggregator would not just output the last timestamp it saw, but instead would output the oldest timestamp it had seen in the past minute across all of its threads (assuming multiple threads potentially reading from multiple partitions). This would give a less "lumpy" view.
This method gives an approximate picture of the end-to-end latency, since it's measuring the point at which the aggregator receives the sales record, not the point at which the database has been updated.
The questions
Is one method more likely to get usable results than the other?
For method 1, is there a sequence of numbers which would be more efficient than powers of 2 in allowing us to work out the earliest value when multiple ones arrive at once, requiring fewer bits so that the time before sequence reset would be longer?
Would method 1 have the same problem of "lumpy" data as method 2, in the case of a large number of partitions or data arriving out of order?
Given that method 2 seems simpler, is the method of smoothing out the lumps in the measurement a plausible one?

Differnces between __execute-count value and values gathered by the Metrics Reporting API v2

I have run a topology, and I used the Meter type in metric Reporting API v2. In the execute method I mark this metric. So it will mark an event whenever the execute method is called. But when I compare this value with the __execute-count, I see huge differences. Does anyone know why this happens?
These are the values from my log which are gathered at the same time:
9:v7 __execute-count {v0:v7=44500}
9:v7 tuple_inRate.count 664129
Update:
When I use the mark method on the Meter metric, I will get different results in comparison with the Counter metric. But still, I do not understand why the values from the counter metric (tuple counter) are not the same as the __execute-count.
As given in this answer, Storms Internal Metrics are just estimated by a percentage of the real data flow. Initially, it uses 5% of incoming tuples to make those estimations. This may lead to inaccuracies for extreme high or low throughputs.
EDIT: The documentation describes the following:
In general all of these tuple count metrics are randomly sub-sampled unless otherwise stated. This means that the counts you see both on the UI and from the built in metrics are not necessarily exact. In fact by default we sample only 5% of the events and estimate the total number of events from that. The sampling percentage is configurable per topology through the topology.stats.sample.rate config. Setting it to 1.0 will make the counts exact, but be aware that the more events we sample the slower your topology will run (as the metrics are counted in the same code path as tuples are processed). This is why we have a 5% sample rate as the default.
EDIT 2 In this post, there is more information about the estimation:
The way it works is that if you choose a sampling rate of 0.05, it will pick a random element of the next 20 events in which to increase the count by 20. So if you have 20 tasks for that bolt, your stats could be off by +-380.
By the way, execute_count is just an increasing number, while your tuple_inRate.count is a rate, isn`t it?

Schedule sending messages to consumers at different rate

I'm looking for best algorithm for message schedule. What I mean with message schedule is a way to send a messages on the bus when we have many consumers at different rate.
Example :
Suppose that we have data D1 to Dn
. D1 to send to many consumer C1 every 5ms, C2 every 19ms, C3 every 30ms, Cn every Rn ms
. Dn to send to C1 every 10ms, C2 every 31ms , Cn every 50ms
What is best algorithm which schedule this actions with the best performance (CPU, Memory, IO)?
Regards
I can think of quite a few options, each with their own costs and benefits. It really comes down to exactly what your needs are -- what really defines "best" for you. I've pseudocoded a couple possibilities below to hopefully help you get started.
Option 1: Execute the following every time unit (in your example, millisecond)
func callEachMs
time = getCurrentTime()
for each datum
for each customer
if time % datum.customer.rate == 0
sendMsg()
This has the advantage of requiring no consistently stored memory -- you just check at each time unit whether your should be sending a message. This can also deal with messages that weren't sent at time == 0 -- just store the time the message was initially sent modulo the rate, and replace the conditional with if time % datum.customer.rate == data.customer.firstMsgTimeMod.
A downside to this method is it is completely reliant on always being called at a rate of 1 ms. If there's lag caused by another process on a CPU and it misses a cycle, you may miss sending a message altogether (as opposed to sending it a little late).
Option 2: Maintain a list of lists of tuples, where each entry represents the tasks that need to be done at that millisecond. Make your list at least as long as the longest rate divided by the time unit (if your longest rate is 50 ms and you're going by ms, your list must be at least 50 long). When you start your program, place the first time a message will be sent into the queue. And then each time you send a message, update the next time you'll send it in that list.
func buildList(&list)
for each datum
for each customer
if list.size < datum.customer.rate
list.resize(datum.customer.rate+1)
list[customer.rate].push_back(tuple(datum.name, customer.name))
func callEachMs(&list)
for each (datum.name, customer.name) in list[0]
sendMsg()
list[customer.rate].push_back((datum.name, customer.name))
list.pop_front()
list.push_back(empty list)
This has the advantage of avoiding the many unnecessary modulus calculations option 1 required. However, that comes with the cost of increased memory usage. This implementation would also not be efficient if there's a large disparity in the rate of your various messages (although you could modify this to deal with algorithms with longer rates more efficiently). And it still has to be called every millisecond.
Finally, you'll have to think very carefully about what data structure you use, as this will make a huge difference in its efficiency. Because you pop from the front and push from the back at every iteration, and the list is a fixed size, you may want to implement a circular buffer to avoid unneeded moving of values. For the lists of tuples, since they're only ever iterated over (random access isn't needed), and there are frequent additions, a singly-linked list may be your best solution.
.
Obviously, there are many more ways that you could do this, but hopefully, these ideas can get you started. Also, keep in mind that the nature of the system you're running this on could have a strong effect on which method works better, or whether you want to do something else entirely. For example, both methods require that they can be reliably called at a certain rate. I also haven't described parallellized implementations, which may be the best option if your application supports them.
Like Helium_1s2 described, there is a second way which based on what I called a schedule table and this is what I used now but this solution has its limits.
Suppose that we have one data to send and two consumer C1 and C2 :
Like you can see we must extract our schedule table and we must identify the repeating transmission cycle and the value of IDLE MINIMUM PERIOD. In fact, it is useless to loop on the smallest peace of time ex 1ms or 1ns or 1mn or 1h (depending on the case) BUT it is not always the best period and we can optimize this loop as follows.
for example one (C1 at 6 and C2 at 9), we remark that there is cycle which repeats from 0 to 18. with a minimal difference of two consecutive send event equal to 3.
so :
HCF(6,9) = 3 = IDLE MINIMUM PERIOD
LCM(6,9) = 18 = transmission cycle length
LCM/HCF = 6 = size of our schedule table
And the schedule table is :
and the sending loop looks like :
while(1) {
sleep(IDLE_MINIMUM_PERIOD); // free CPU for idle min period
i++; // initialized at 0
send(ScheduleTable[i]);
if (i == sizeof(ScheduleTable)) i=0;
}
The problem with this method is that this array will grows if LCM grows which is the case if we have bad combination like with rate = prime number, etc.

Apache Storm: Limit the number of tuples produced (in a certain time frame)

By my understanding ack, fail and nextTuple functions are called using the same thread in a tight loop as ISpout Javadoc suggest:
nextTuple, ack, and fail are all called in a tight loop in a single
thread in the spout task.
Say we have a synthetic tuple generator and we want to limit the number of tuples emitted by the spout per second. How can that be achieved? Is putting a sleep() a good idea? Is there an alternative way?
A sleep might not be the best idea because it blocks the spout to process incoming acks. See here: Why should I not loop or block in Spout.nextTuple()
I would just count the emitted tuples and remember a timestamp. If the number of tuples per time unit is exceeded and the time unit did not pass, just return from nextTuple() without emitting any tuple. When the time unit passed, reset the counter to zero, advance the timestamp by the time unit, and resume emitting. And so fourth.

Resources