How to forward to the original uri in the balancer.lua, when it have change in the plugin?

[xianshun163]

Issue description

Dear all, I use the proxy-rewrite to change the uri, then in the balancer.lua, if all the backupend server is avalable, I want to forwards the request to other cluster , so I want to set the uri to the original, but I can not do it.
I try it like this : ctx.var.upstream_uri = ctx.old_upstream_uri
but it not work now, It still forwards the modified URI, not the original.

Environment

• apisix version (cmd: apisix version):1.11.0
• OS (cmd: uname -a):center os 7
• OpenResty / Nginx version (cmd: nginx -V or openresty -V):
• etcd version, if have (cmd: run curl http://127.0.0.1:9090/v1/server_info to get the info from server-info API):
• apisix-dashboard version, if have:
• the plugin runner version, if the issue is about a plugin runner (cmd: depended on the kind of runner):
• luarocks version, if the issue is about installation (cmd: luarocks --version):

[xianshun163]

In other words, in the balancer phase, can I still modify the URI?

[tzssangglass]

1. you need rto use balancer.recreate_request after you changed the uri;
2. you can hook before_proxy in plugins to do this, see:

   apisix/apisix/balancer.lua

   Lines 342 to 346 in b52f95b

   	local _, run = plugin_funcs("before_proxy")
   	-- always recreate request as the request may be changed by plugins
   	if (run or header_changed) and balancer.recreate_request then
   	balancer.recreate_request()
   	end

   , then you don't need to call balancer.recreate_request

[xianshun163]

@tzssangglass ,

2. balancer.recreate_request()

Thank you for your reply. I don't fully understand your content.
I let the balancer.recreate_ Request () at balancer.Lua code 345 line , is called or not, and the URI is still the same as proxy_ rewrite. Lua has been modified, without restoring the original URI. Is my test wrong?

In addition, my purpose is to work in balancer.Lua, when all the upstream unavailable, the request is transferred to another backup gateway , so the URI should be restored to its original appearance.

[tzssangglass]

I just remind you that you need balancer.recreate_request() after you modify the uri in the balancer phase.

Can you show how you modified it?

[xianshun163]

I modify it like this:
1、in balancer.lua local function pick_server(route, ctx)
-- when all node is fail
ctx.balancer_server = my_backup_gateway_server
ctx.var.upstream_uri = ctx.var._cache.real_request_uri
ctx.donot_change_uri = true
2、in balancer.lua function _M.run(route, ctx, plugin_funcs)
if (run or header_changed or ctx.donot_change_uri ) and balancer.recreate_request then
core.log.warn("do recreate_request")
balancer.recreate_request()
end

[tzssangglass]

if (run or header_changed or ctx.donot_change_uri ) and balancer.recreate_request then
core.log.warn("do recreate_request")
balancer.recreate_request()
end

This code will only be executed during the balancer retry phase.

I've thought about it for a while, and your needs are a bit specific and a bit tricky to implement.

I'm turning this issue into a discussion because it's not related to the project.

[xianshun163]

if (run or header_changed or ctx.donot_change_uri ) and balancer.recreate_request then
core.log.warn("do recreate_request")
ctx.var.upstream_uri = "/test/change/uri"
balancer.recreate_request()
end

yes, the code is executed, but it can not change the uri yet.

[tzssangglass]

yes, simply put, this code was not executed at the right time.

I've thought of a few points to check:

1. When is it confirmed that all the backup end server is avalable?

The backup end server will be attempted only if the high-priority end server is unavailable. You need to cut in when you get to the server_picker selection to the backup end server.

You can take a look about:

apisix/apisix/balancer.lua

Line 117 in 682f8cc

local server_picker = priority_balancer.new(up_nodes, upstream, picker)