8.3 KiB
Thread pool
The thread pool is a working server, made of a set of worker threads that can be mapped on CPU cores.
Each worker loop on pick from the same input queue jobs to do.
When a job is done, the worker sends a return if a return is defined.
A selective abort allows to cancel parallel jobs (usage: a client pushed jobs, but from a response of one job, the other linked jobs becomes useless).
All the thread pool functions are thread safe, nevertheless the working functions are implemented by the thread pool client, so the client has to tackle the parallel execution of his functions called "processingFunc" hereafter.
license
Author:
Laurent Thomas, Open cells project
The owner share this piece code to Openairsoftware alliance as per OSA license terms
jobs
A job is a message (notifiedFIFO_elt_t):
- next: internal FIFO chain, do not set it
- key: a long int that the client can use to identify a message or a group of messages
- ResponseFifo: if the client defines a response FIFO, the message will be posted back after processing
- processingFunc: any funtion (type void processingFunc(void *)) that the worker will launch
- msgData: the data passed to processingFunc. It can be added automatically, or you can set it to a buffer you are managing
- malloced: a boolean that enable internal free in these cases: no return Fifo or Abort feature
The job messages can be created with newNotifiedFIFO_elt() and delNotifiedFIFO_elt() or managed by the client.
Queues of jobs
Queues are type of:
- notifiedFIFO_t that must be initialized by init_notifiedFIFO()
- No delete function is required, the creator has only to free the data of type notifiedFIFO_t
- push_notifiedFIFO() add a job in the queue
- pull_notifiedFIFO() is blocking, poll_notifiedFIFO() is non blocking
- abort_notifiedFIFO() allows the customer to delete all waiting jobs that match with the key (see key in jobs definition)
These queues details hereafter
Common
newNotifiedFIFO_elt()
creates a message, that will later be used in queues/FIFO
delNotifiedFIFO_elt()
deletes it
NotifiedFifoData()
gives a pointer to the beginning of free usage memory in a message (you can put any data there, up to 'size' parameter you passed to newNotifiedFIFO_elt()
These 3 calls are not mandatory, you can also use your own memory to save the malloc()/free() that are behind these calls iFirst level: non thread safe FIFO (or queues)
low level: fast and simple
initNotifiedFIFO_nothreadSafe()
to create a queue
pushNotifiedFIFO_nothreadSafe()
Add a element in a queue
pullNotifiedFIFO_nothreadSafe()
pull a element from a queue
As these queues are not thread safe, there is NO blocking mechanism, neither pull() versus poll() calls
There is no delete for a message queue: you only have to abandon the memory you allocated to call iinitNotifiedFIFO_nothreadSafe(notifiedFIFO_t *nf)
So if you malloced the memory under 'nf' parameter you have to free it, if it is automatic variable (local variable) or global variable, nothing is to be done.
thread safe queues
These queues are built on not thread safe queues when we need thread to thread protection
initNotifiedFIFO()
to create a queue
pushNotifiedFIFO()
Add a element in a queue
pullNotifiedFIFO()
pull a element from a queue, this call is blocking until a message arrived
pollNotifiedFIFO()
pull a element from a queue, this call is not blocking, so it returns always very shortly
in 99.9% cases, pull() is better than poll()
No delete() call, same principle as not thread safe queues
Thread pools
initialization
The clients can create one or more thread pools with initTpool(char *params,tpool_t *pool, bool performanceMeas ) or initNamedTpool(char *params,tpool_t *pool, bool performanceMeas , char *name)
the params string structure: describes a list of cores, separated by "," that run a worker thread
If the core exists on the CPU, the thread pool initialization sets the affinity between this thread and the related code (use negative values is allowed, so the thread will never be mapped on a specific core).
The threads are all Linux real time scheduler, their name is set automatically to Tpool<thread index>_<core id> if initTpool is used or to <name><thread index>_<core id> when initNamedTpool is used.
adding jobs
The client create their jobs messages as a notifiedFIFO_elt_t, then they push it with pushTpool() (that internally calls push_notifiedFIFO())
If they need a return, they have to create response queues with init_notifiedFIFO() and set this FIFO pointer in the notifiedFIFO_elt_t before pushing the job.
abort
A abort service abortTpool() allows to abort all jobs that match a key (see jobs "key"). When the abort returns, it garanties no job (matching the key) response will be posted on response queues.
Nevertheless, jobs already performed before the return of abortTpool() are pushed in the response Fifo queue.
API details
Each thread pool (there can be several in the same process) should be initialized once using one of the two API's:
initNamedTpool(char *params,tpool_t *pool, bool performanceMeas,char *name)
initTpool(char *params,tpool_t *pool, bool performanceMeas)
params: the configuration parameter is a string, elements separator is a comma ",". An element can be:
- N: if a N is in the parameter list, no threads are created The purpose is to keep the same API in any case
- a CPU with a little number of cores, or in debugging sessions to simplify the human work
- a number that represent a valid CPU core on the target CPU A thread is created and stick on the core (with set affinity)
- a number that is not a valid CPU core a floating thread is created (Linux is responsible of the real time core allocation)
example: "-1,-1,-1" as there is no core number -1, the thread pool is made of 3 floating threads be careful with fix allocation: it is hard to be more clever than Linux kernel
pool is the memory you allocated before to store the thread pool internal state (same concept as above queues)
performanceMeas is a flag to enable measurements (well described in documentation)
name is used to build the thread names.
pushTpool(tpool_t *t, notifiedFIFO_elt_t *msg)
adds a job to do in the thread pool
The msg data you can set are:
- key:
a value for you that you will find back in the response it is also the key for abortTpool() - reponseFifo
if you set it, the message will be sent back on this queue when the job is done if you don't set it, no return is performed, the thread pool frees the message 'msg' when the job is done - processingFunc
the function the job will run. the function prototype is void (void *memory) the data part (the pointer returned by NotifiedFifoData(msg)) is passed to the function it is used to send data to the processing function and also to write back results of course, writing back results will lead you to use also a return queue (the parameter reponseFifo)
pullTpool() collects job result in a return queue
you collect results in one result queue: the message you gave to pushTpool(), nevertheless it has been updated by processingFunc()
An example of multiple return queues, in eNB: I created one single thread pool (because it depends mainly on CPU hardware), but i use two return queues: one for turbo encode, one for turbo decode.
tryPullTpool()
is the same, but not blocking (pollTpool() would have been a better name)
abortTpool()
Is a facility to cancel work you pushed to a thread pool
I used it once: when eNB performs turbo decode, I push all segments in the thread pool.
But when I get back the decoding results, if one segment can't be decoded, I don't need the results of the other segments of the same UE.
Performance measurements
A performance measurement is integrated: the pool will automacillay fill timestamps:
- creationTime: time the request is push to the pool;
- startProcessingTime: time a worker start to run on the job
- endProcessingTime: time the worker finished the job
- returnTime: time the client reads the result
if you set the environement variable: threadPoolMeasurements to a valid file name These measurements will be wrote to this Linux pipe.
A tool to read the linux fifo and display it in ascii is provided; compute the binay: in cmake building directory: make/ninja measurement_display