You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@apisix.apache.org by Jan Li <ja...@airwallex.com> on 2020/05/08 11:43:49 UTC
[DISCUSS] APISIX should fail fast if mandatory dependency is not available
Hi team,
When using APISIX in critical env, we'd like it fail fast to avoid
false healthy server which cannot process any actual requests:
1. If APISIX cannot connect to ETCD when it's configured with the
ETCD config center, it means pre-configured routes are not available. At
this point, APISIX should fail the start, but currently it starts as if
everything is fine, but since all preconfigured routes are not available,
it cannot serve any real business requests.
2. If a plugin has a critical init phase which should be completed
before the plugin can be used in request processing, and the init phase
throws an error. APISIX should fail at the start in such cases, since it
cannot process any real business requests which have that fail-to-init
plugin configured.
For the first, I think APISIX should verify ETCD connectivity if
configured to use ETCD as the config center.
For the second, I think every plugin should be able to declare some
critical method which must be completed before the plugin is ready to serve
requests, such as a main_init() method which will be called inside
init_by_lua_block phase. If such a plugin fails to init itself, APISIX
should not start itself.
Re: [DISCUSS] APISIX should fail fast if mandatory dependency is not available
Posted by YuanSheng Wang <me...@apache.org>.
We can add the detection step in the CLI tool.
1. Use the curl tool to check if the etcd service is working properly
2. Start the apache apisix service.
On Fri, May 8, 2020 at 10:06 PM Ming Wen <we...@apache.org> wrote:
> Hi,Jan,
> Thanks for the advice, I totally agree with you.
> It's much better to report an error loudly than silently.
>
> For the first one, we did a test[1] but there is no error and no
> stop-start,
> it's an error, feel free to submit a PR to fix it.
>
> For the second, the plugin's init function does not actually run in nginx's
> init_by_lua block,
> But we can also consider other ways to handle.
>
> [1] https://github.com/apache/incubator-apisix/blob/master/bin/apisix#L801
>
> Thanks,
> Ming Wen, Apache APISIX & Apache SkyWalking
> Twitter: _WenMing
>
>
> Jan Li <ja...@airwallex.com> 于2020年5月8日周五 下午7:44写道：
>
> > Hi team,
> >
> > When using APISIX in critical env, we'd like it fail fast to avoid
> > false healthy server which cannot process any actual requests:
> >
> > 1. If APISIX cannot connect to ETCD when it's configured with the
> > ETCD config center, it means pre-configured routes are not available. At
> > this point, APISIX should fail the start, but currently it starts as if
> > everything is fine, but since all preconfigured routes are not available,
> > it cannot serve any real business requests.
> > 2. If a plugin has a critical init phase which should be completed
> > before the plugin can be used in request processing, and the init phase
> > throws an error. APISIX should fail at the start in such cases, since it
> > cannot process any real business requests which have that fail-to-init
> > plugin configured.
> >
> > For the first, I think APISIX should verify ETCD connectivity if
> > configured to use ETCD as the config center.
> >
> > For the second, I think every plugin should be able to declare some
> > critical method which must be completed before the plugin is ready to
> serve
> > requests, such as a main_init() method which will be called inside
> > init_by_lua_block phase. If such a plugin fails to init itself, APISIX
> > should not start itself.
> >
>
--
*MembPhis*
My github: https://github.com/membphis
Apache APISIX: https://github.com/apache/incubator-apisix
Re: [DISCUSS] APISIX should fail fast if mandatory dependency is not available
Posted by Ming Wen <we...@apache.org>.
Hi,Jan,
Thanks for the advice, I totally agree with you.
It's much better to report an error loudly than silently.
For the first one, we did a test[1] but there is no error and no stop-start,
it's an error, feel free to submit a PR to fix it.
For the second, the plugin's init function does not actually run in nginx's
init_by_lua block,
But we can also consider other ways to handle.
[1] https://github.com/apache/incubator-apisix/blob/master/bin/apisix#L801
Thanks,
Ming Wen, Apache APISIX & Apache SkyWalking
Twitter: _WenMing
Jan Li <ja...@airwallex.com> 于2020年5月8日周五 下午7:44写道：
> Hi team,
>
> When using APISIX in critical env, we'd like it fail fast to avoid
> false healthy server which cannot process any actual requests:
>
> 1. If APISIX cannot connect to ETCD when it's configured with the
> ETCD config center, it means pre-configured routes are not available. At
> this point, APISIX should fail the start, but currently it starts as if
> everything is fine, but since all preconfigured routes are not available,
> it cannot serve any real business requests.
> 2. If a plugin has a critical init phase which should be completed
> before the plugin can be used in request processing, and the init phase
> throws an error. APISIX should fail at the start in such cases, since it
> cannot process any real business requests which have that fail-to-init
> plugin configured.
>
> For the first, I think APISIX should verify ETCD connectivity if
> configured to use ETCD as the config center.
>
> For the second, I think every plugin should be able to declare some
> critical method which must be completed before the plugin is ready to serve
> requests, such as a main_init() method which will be called inside
> init_by_lua_block phase. If such a plugin fails to init itself, APISIX
> should not start itself.
>