Module: sip-router Branch: master Commit: c92ec42ac07a4a000a95dcd1f3c1a349ce35ea82 URL: http://git.sip-router.org/cgi-bin/gitweb.cgi/sip-router/?a=commit;h=c92ec42a...
Author: Daniel-Constantin Mierla miconda@gmail.com Committer: Daniel-Constantin Mierla miconda@gmail.com Date: Mon Apr 14 22:37:02 2014 +0200
acc: extended documentation for db_insert_mode parameter
- if set to 2, async insert operation is used to store the acc record in database table
---
modules/acc/README | 3 +++ modules/acc/doc/acc_admin.xml | 4 ++++ 2 files changed, 7 insertions(+), 0 deletions(-)
diff --git a/modules/acc/README b/modules/acc/README index da2df70..bf53f25 100644 --- a/modules/acc/README +++ b/modules/acc/README @@ -1113,6 +1113,9 @@ modparam("acc", "db_extra", "ct=$hdr(Content-type); email=$avp(s:email)") InnoDB engine doesn't support INSERT DELAYED, thus be sure the acc tables are defined with different type (e.g., MyISAM).
+ If set to 2, async insert is used if the db driver module has support + for it. If not, then standard INSERT is used. + Default value is 0 (no INSERT DELAYED).
Example 1.32. db_insert_mode example diff --git a/modules/acc/doc/acc_admin.xml b/modules/acc/doc/acc_admin.xml index c59aa0e..8b5ee60 100644 --- a/modules/acc/doc/acc_admin.xml +++ b/modules/acc/doc/acc_admin.xml @@ -1098,6 +1098,10 @@ modparam("acc", "db_extra", "ct=$hdr(Content-type); email=$avp(s:email)") the acc tables are defined with different type (e.g., MyISAM). </para> <para> + If set to 2, async insert is used if the db driver module has + support for it. If not, then standard INSERT is used. + </para> + <para> Default value is 0 (no INSERT DELAYED). </para> <example>
Daniel-Constantin Mierla writes:
- If set to 2, async insert is used if the db driver module has support
- for it. If not, then standard INSERT is used.
- Default value is 0 (no INSERT DELAYED).
daniel,
just to clarity, is there some relationship between INSERT DELAYED and async insert? if db_insert_mode=2, is INSERT DELAYED or standard INSERT used? that is, do db_insert_mode 0 and 2 somehow differ on the wire to db server or is it so that both 0 and 2 use standard INSERT?
-- juha
Hello,
On 16/04/14 10:48, Juha Heinanen wrote:
Daniel-Constantin Mierla writes:
- If set to 2, async insert is used if the db driver module has support
- for it. If not, then standard INSERT is used.
- Default value is 0 (no INSERT DELAYED).
daniel,
just to clarity, is there some relationship between INSERT DELAYED and async insert?
No.
if db_insert_mode=2, is INSERT DELAYED or standard INSERT used? that is, do db_insert_mode 0 and 2 somehow differ on the wire to db server or is it so that both 0 and 2 use standard INSERT?
Both use standard INSERT.
INSERT DELAYED is an extension of MySQL and doesn't work for all engines (e.g., innodb) and there are even plans to be removed, as I got it from some forums.
The async insert is based on a generic framework that can be extended to all other queries that don't return a db result needed in config or inside the module (e.g., delete, update -- on the todo list, trying to figure out if would be possible to do it via db api for all db modules at once).
In the recent commits I added the features of creating a dedicated group of processes (named async workers) that listen on an internal socket for tasks sent by other processes (e.g., sip workers, timers, etc.). These new processes are used also by the async_task_route() function.
The mechanism behind async insert is as follow: - the INSERT query is printed by the process that does the db operation - the mysql module instead of sending the query to mysql server, clones it in shared memory and sends it to the group of async workers - one of the workers reads the task and sends the query to the database server
To have the above working, the db_url is also cloned and sent to async workers in order to identify the right connection to use. The first idle async worker is receiving the task (it is kernel based, the same mechanism used for many processes reading on the same UDP socket). You have to specify the number of async workers via global parameter async_workers:
- http://www.kamailio.org/wiki/cookbooks/devel/core#async_workers
Cheers, Daniel
Daniel-Constantin Mierla writes:
Default value is 0 (no INSERT DELAYED).daniel,
just to clarity, is there some relationship between INSERT DELAYED and async insert?
No.
good. so perhaps the comment on default value could be (no INSERT DELAYED nor async insert is used).
The mechanism behind async insert is as follow:
- the INSERT query is printed by the process that does the db operation
- the mysql module instead of sending the query to mysql server, clones
it in shared memory and sends it to the group of async workers
- one of the workers reads the task and sends the query to the database
server
does there need to be a free async worker to serve the request or are the requests buffered so that the sender of the request is always immediately free to go on with other business?
To have the above working, the db_url is also cloned and sent to async workers in order to identify the right connection to use. The first idle async worker is receiving the task (it is kernel based, the same mechanism used for many processes reading on the same UDP socket). You have to specify the number of async workers via global parameter async_workers:
that would have been my next question. if someone only reads acc/README, it does not mention a need set async_workers > 0 in order to benefit from db_insert_mode=2.
-- juha
On 16/04/14 11:24, Juha Heinanen wrote:
Daniel-Constantin Mierla writes:
Default value is 0 (no INSERT DELAYED).daniel,
just to clarity, is there some relationship between INSERT DELAYED and async insert?
No.
good. so perhaps the comment on default value could be (no INSERT DELAYED nor async insert is used).
You can enhance the docs as you consider more clear -- the possible values are listed in the same section.
The mechanism behind async insert is as follow:
- the INSERT query is printed by the process that does the db operation
- the mysql module instead of sending the query to mysql server, clones
it in shared memory and sends it to the group of async workers
- one of the workers reads the task and sends the query to the database
server
does there need to be a free async worker to serve the request or are the requests buffered so that the sender of the request is always immediately free to go on with other business?
The sender is immediately free. The request is queued in the internal socket (used to pass only the pointer to shared memory query) and will be handled by first async worker. In other words, the sender does like an 'udp send', but via a reliable socket.
To have the above working, the db_url is also cloned and sent to async workers in order to identify the right connection to use. The first idle async worker is receiving the task (it is kernel based, the same mechanism used for many processes reading on the same UDP socket). You have to specify the number of async workers via global parameter async_workers:
that would have been my next question. if someone only reads acc/README, it does not mention a need set async_workers > 0 in order to benefit from db_insert_mode=2.
You can add a note about this as well, if you put more content in the docs for this parameter.
Cheers, Daniel
-
Daniel-Constantin Mierla -
Juha Heinanen