You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@apisix.apache.org by 聂永 <ni...@staff.weibo.com> on 2021/04/14 09:45:40 UTC
[DISCUSS] dump etcd data and read it when required
Hi,
We need the `etcd` cluster for storing gateway's config, but we are afraid of it being inaccessible when we reload or deploy gateway distance as network issues.
## Dump etcd
So, we need the apisix to dump the etcd's stored data into `conf/apisix.yaml ` file, when it is running every 5 minutes.
And, if the already existed apisix distance need to restart in some unexpected case, or we need do deploy a large number of distance online in a short time, if the distance queries data from `etcd` unsuccessfully when starting, we need to load the dumped `conf/apisix.yaml` file into memory once.
Firstly, we need to import the `dump` for `etcd` config:
```yaml
etcd:
host:
- "http://172.19.5.10:23792"
prefix: "/apisix"
dump:
interval: 300 # unit seconds
load_on_failure: true # if true, then try to load data from dumped conf/apisix.yaml file when querying etcd failure first
```
And, when initializing, it will start a timer for dumping in cycles:
```lua
local dump_conf = etcd_conf.dump
local dump_interval = dump_conf.interval or 300
if dump_interval > 0 and 0 == ngx.worker.id() then
ngx_timer_every(dump_interval, dump_etcd)
end
```
If it queries `etcd` failure first time, then will try to load data from `conf/apisix.yaml` file once,(with its work flow pic https://user-images.githubusercontent.com/548385/114687020-e8dc7b80-9d45-11eb-94ea-1f28465e0580.png), and its code snippet below:
```lua
local ok, err = sync_data(self)
if err then
if i == 1 and self.fail_load_yaml and not failure_map[self.key] then
failure_map[self.key] = 1
try_load_data_from_yaml(self)
end
......
```
## More usage
If we use [`Stand-alone mode`](https://github.com/apache/apisix/blob/master/docs/en/latest/stand-alone.md) below:
```yaml
config_center: yaml
```
We can also use it to update `conf/apisix.yaml` file periodically.
Re: [DISCUSS] dump etcd data and read it when required
Posted by Zexuan Luo <sp...@apache.org>.
Note that the config run after filter will be different from the raw
input. So if dump is enabled, we should create a copy of the original
data and dump this one.
https://github.com/apache/apisix/blob/cd4d2ece58a1075545d3a2dc929b014049f53f21/apisix/upstream.lua#L470
聂永 <ni...@staff.weibo.com> 于2021年4月14日周三 下午5:46写道：
>
> Hi,
>
> We need the `etcd` cluster for storing gateway's config, but we are afraid of it being inaccessible when we reload or deploy gateway distance as network issues.
>
> ## Dump etcd
>
> So, we need the apisix to dump the etcd's stored data into `conf/apisix.yaml ` file, when it is running every 5 minutes.
>
> And, if the already existed apisix distance need to restart in some unexpected case, or we need do deploy a large number of distance online in a short time, if the distance queries data from `etcd` unsuccessfully when starting, we need to load the dumped `conf/apisix.yaml` file into memory once.
>
> Firstly, we need to import the `dump` for `etcd` config:
>
> ```yaml
> etcd:
> host:
> - "http://172.19.5.10:23792"
> prefix: "/apisix"
> dump:
> interval: 300 # unit seconds
> load_on_failure: true # if true, then try to load data from dumped conf/apisix.yaml file when querying etcd failure first
> ```
>
> And, when initializing, it will start a timer for dumping in cycles:
>
> ```lua
> local dump_conf = etcd_conf.dump
> local dump_interval = dump_conf.interval or 300
>
> if dump_interval > 0 and 0 == ngx.worker.id() then
> ngx_timer_every(dump_interval, dump_etcd)
> end
> ```
>
> If it queries `etcd` failure first time, then will try to load data from `conf/apisix.yaml` file once,(with its work flow pic https://user-images.githubusercontent.com/548385/114687020-e8dc7b80-9d45-11eb-94ea-1f28465e0580.png), and its code snippet below:
>
> ```lua
> local ok, err = sync_data(self)
> if err then
> if i == 1 and self.fail_load_yaml and not failure_map[self.key] then
> failure_map[self.key] = 1
> try_load_data_from_yaml(self)
> end
>
> ......
> ```
>
> ## More usage
>
> If we use [`Stand-alone mode`](https://github.com/apache/apisix/blob/master/docs/en/latest/stand-alone.md) below:
>
> ```yaml
> config_center: yaml
> ```
>
> We can also use it to update `conf/apisix.yaml` file periodically.
>
>
>
>
Re: [DISCUSS] dump etcd data and read it when required
Posted by 聂永 <ni...@staff.weibo.com>.
Hi,
You can choose only one distance to dump the ETCD data, just define a rational value for `etcd/dump/interval`:
```yaml
etcd:
...
dump:
interval: 300
```
And you can also set the others to disable the dump action:
```yaml
etcd:
...
dump:
interval: 0
load_on_failure: true
```
2021年4月15日 10:05，Chao Zhang <zc...@gmail.com>> 写道：
外部邮件，慎点链接。
Hi!
As this is an event with low probabilities, I think the dump function can
be implemented in a separate script, when the network is partitioned, running
this script (maybe in one of the pod/vm of etcd instances) and getting this config.yaml,
uploading/mount it to the machine/pod of APISIX. APISIX only needs to support load
the static configuration when communicating to etcd failed.
Chao Zhang
https://github<https://github/>.com/tokers
On April 14, 2021 at 17:46:04, 聂永 (nieyong@staff.weibo.com<ma...@staff.weibo.com>) wrote:
Hi,
We need the `etcd` cluster for storing gateway's config, but we are afraid of it being inaccessible when we reload or deploy gateway distance as network issues.
## Dump etcd
So, we need the apisix to dump the etcd's stored data into `conf/apisix.yaml ` file, when it is running every 5 minutes.
And, if the already existed apisix distance need to restart in some unexpected case, or we need do deploy a large number of distance online in a short time, if the distance queries data from `etcd` unsuccessfully when starting, we need to load the dumped `conf/apisix.yaml` file into memory once.
Firstly, we need to import the `dump` for `etcd` config:
```yaml
etcd:
host:
- "http://172.19.5.10:23792<http://172.19.5.10:23792/>"
prefix: "/apisix"
dump:
interval: 300 # unit seconds
load_on_failure: true # if true, then try to load data from dumped conf/apisix.yaml file when querying etcd failure first
```
And, when initializing, it will start a timer for dumping in cycles:
```lua
local dump_conf = etcd_conf.dump
local dump_interval = dump_conf.interval or 300
if dump_interval > 0 and 0 == ngx.worker.id<http://ngx.worker.id/>() then
ngx_timer_every(dump_interval, dump_etcd)
end
```
If it queries `etcd` failure first time, then will try to load data from `conf/apisix.yaml` file once,(with its work flow pichttps://user-images.githubusercontent.com/548385/114687020-e8dc7b80-9d45-11eb-94ea-1f28465e0580.png), and its code snippet below:
```lua
local ok, err = sync_data(self)
if err then
if i == 1 and self.fail_load_yaml and not failure_map[self.key] then
failure_map[self.key] = 1
try_load_data_from_yaml(self)
end
......
```
## More usage
If we use [`Stand-alone mode`](https://github.com/apache/apisix/blob/master/docs/en/latest/stand-alone.md) below:
```yaml
config_center: yaml
```
We can also use it to update `conf/apisix.yaml` file periodically.
Re: [DISCUSS] dump etcd data and read it when required
Posted by Chao Zhang <zc...@gmail.com>.
Hi!
As this is an event with low probabilities, I think the dump function can
be implemented in a separate script, when the network is partitioned,
running
this script (maybe in one of the pod/vm of etcd instances) and getting this
config.yaml,
uploading/mount it to the machine/pod of APISIX. APISIX only needs to
support load
the static configuration when communicating to etcd failed.
Chao Zhang
https://github.com/tokers
On April 14, 2021 at 17:46:04, 聂永 (nieyong@staff.weibo.com) wrote:
Hi,
We need the `etcd` cluster for storing gateway's config, but we are afraid
of it being inaccessible when we reload or deploy gateway distance as
network issues.
## Dump etcd
So, we need the apisix to dump the etcd's stored data into
`conf/apisix.yaml ` file, when it is running every 5 minutes.
And, if the already existed apisix distance need to restart in some
unexpected case, or we need do deploy a large number of distance online in
a short time, if the distance queries data from `etcd` unsuccessfully when
starting, we need to load the dumped `conf/apisix.yaml` file into memory
once.
Firstly, we need to import the `dump` for `etcd` config:
```yaml
etcd:
host:
- "http://172.19.5.10:23792"
prefix: "/apisix"
dump:
interval: 300 # unit seconds
load_on_failure: true # if true, then try to load data from dumped
conf/apisix.yaml file when querying etcd failure first
```
And, when initializing, it will start a timer for dumping in cycles:
```lua
local dump_conf = etcd_conf.dump
local dump_interval = dump_conf.interval or 300
if dump_interval > 0 and 0 == ngx.worker.id() then
ngx_timer_every(dump_interval, dump_etcd)
end
```
If it queries `etcd` failure first time, then will try to load data from
`conf/apisix.yaml` file once,(with its work flow pic
https://user-images.githubusercontent.com/548385/114687020-e8dc7b80-9d45-11eb-94ea-1f28465e0580.png),
and its code snippet below:
```lua
local ok, err = sync_data(self)
if err then
if i == 1 and self.fail_load_yaml and not failure_map[self.key] then
failure_map[self.key] = 1
try_load_data_from_yaml(self)
end
......
```
## More usage
If we use [`Stand-alone mode`](
https://github.com/apache/apisix/blob/master/docs/en/latest/stand-alone.md)
below:
```yaml
config_center: yaml
```
We can also use it to update `conf/apisix.yaml` file periodically.