[
https://issues.apache.org/jira/browse/TS-1740?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yunkai Zhang updated TS-1740:
-----------------------------
Description:
Improve precision of stats's values
Conversion between types, such as RecInt(int64_t) to RecFloat(float),
may lead to loss of precision.
In the old code, token value was defined with StatFloat type, it would casue
unnecessary conversion when type of token is not StatFloat.
it's best to keep value with the same type of token as much as possiable.
===
For example, we have defined such expression in stats.config.xml:
{code}
<statistics>
<destination>proxy.node.http.user_agent_total_request_bytes</destination>
<destination
scope="cluster">proxy.cluster.http.user_agent_total_request_bytes</destination>
<expression>
proxy.process.http.user_agent_request_document_total_size +
proxy.process.http.user_agent_request_header_total_size
</expression>
</statistics>
{code}
So:
proxy.node.http.user_agent_total_request_bytes = \
proxy.process.http.user_agent_request_document_total_size + \
proxy.process.http.user_agent_request_header_total_size
These three stats token are all RecInt types, but the code would convert them
to RecFloat when evaluating the expression, like that:
proxy.node.http.user_agent_total_request_bytes = \
(RecFloat)proxy.process.http.user_agent_request_document_total_size + \
(RecFloat)proxy.process.http.user_agent_request_header_total_size
As a result, we might get the following output from stats http API:
{code}
[root@test62 trafficserver]# cat ~/bin/list-cluster.sh
elinks -dump http://localhost:8080/stat/ | grep proxy_name
elinks -dump http://localhost:8080/stat/ | grep nodes
elinks -dump http://localhost:8080/stat/ | grep
proxy.process.http.origin_server_request_document_total_size
elinks -dump http://localhost:8080/stat/ | grep
proxy.process.http.origin_server_request_header_total_size
elinks -dump http://localhost:8080/stat/ | grep
proxy.node.http.origin_server_total_request_bytes
[root@test62 trafficserver]# list-cluster.sh
proxy.config.proxy_name=test62.63
proxy.node.cluster.nodes=1
proxy.process.cluster.nodes=1
proxy.process.http.origin_server_request_document_total_size=0 <= #1
proxy.process.http.origin_server_request_header_total_size=359209123 <= #2
proxy.node.http.origin_server_total_request_bytes=359209120 <= #3 != (#1 + #2)
{code}
was:
Improve precision of stats's values
Conversion between types, such as RecInt(int64_t) to RecFloat(float),
may lead to loss of precision.
In the old code, token value was defined with StatFloat type, it would casue
unnecessary conversion when type of token is not StatFloat.
it's best to keep value with the same type of token as much as possiable.
> Improve precision of stats's values
> -----------------------------------
>
> Key: TS-1740
> URL: https://issues.apache.org/jira/browse/TS-1740
> Project: Traffic Server
> Issue Type: Bug
> Components: Stats
> Reporter: Yunkai Zhang
> Attachments: 0001-Improve-precision-of-stats-s-values.patch
>
>
> Improve precision of stats's values
> Conversion between types, such as RecInt(int64_t) to RecFloat(float),
> may lead to loss of precision.
> In the old code, token value was defined with StatFloat type, it would casue
> unnecessary conversion when type of token is not StatFloat.
> it's best to keep value with the same type of token as much as possiable.
> ===
> For example, we have defined such expression in stats.config.xml:
> {code}
> <statistics>
>
> <destination>proxy.node.http.user_agent_total_request_bytes</destination>
> <destination
> scope="cluster">proxy.cluster.http.user_agent_total_request_bytes</destination>
> <expression>
> proxy.process.http.user_agent_request_document_total_size +
> proxy.process.http.user_agent_request_header_total_size
> </expression>
> </statistics>
> {code}
> So:
> proxy.node.http.user_agent_total_request_bytes = \
> proxy.process.http.user_agent_request_document_total_size + \
> proxy.process.http.user_agent_request_header_total_size
> These three stats token are all RecInt types, but the code would convert them
> to RecFloat when evaluating the expression, like that:
> proxy.node.http.user_agent_total_request_bytes = \
> (RecFloat)proxy.process.http.user_agent_request_document_total_size + \
> (RecFloat)proxy.process.http.user_agent_request_header_total_size
> As a result, we might get the following output from stats http API:
> {code}
> [root@test62 trafficserver]# cat ~/bin/list-cluster.sh
> elinks -dump http://localhost:8080/stat/ | grep proxy_name
> elinks -dump http://localhost:8080/stat/ | grep nodes
> elinks -dump http://localhost:8080/stat/ | grep
> proxy.process.http.origin_server_request_document_total_size
> elinks -dump http://localhost:8080/stat/ | grep
> proxy.process.http.origin_server_request_header_total_size
> elinks -dump http://localhost:8080/stat/ | grep
> proxy.node.http.origin_server_total_request_bytes
> [root@test62 trafficserver]# list-cluster.sh
> proxy.config.proxy_name=test62.63
> proxy.node.cluster.nodes=1
> proxy.process.cluster.nodes=1
> proxy.process.http.origin_server_request_document_total_size=0 <= #1
> proxy.process.http.origin_server_request_header_total_size=359209123 <= #2
> proxy.node.http.origin_server_total_request_bytes=359209120 <= #3 != (#1 +
> #2)
> {code}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
