最近一直在开发自动化运维发布平台，底层命令行、文件通道主要基于paramiko模块，使用过程中遇到各种各样的问题，本文主要用于收集问题及解决记录，以备后续使用。
一、Error reading SSH protocol banner连接错误
这个关键词，在百度、谷歌一搜一大把的提问，也有少部分给出了解决方案，但是最终都无法解决，我经过不断尝试和解读paramiko源码，终于搞定了这个问题，在此记录分享下。
1、具体报错信息：
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "build/bdist.linux-x86_64/egg/paramiko/client.py", line 307, in connect
File "build/bdist.linux-x86_64/egg/paramiko/transport.py", line 465, in start_client
paramiko.SSHException: Error reading SSH protocol banner
2、解决办法：
重新下载paramiko插件源码，解压后，编辑安装目录下的transport.py文件：
vim build/lib/paramiko/transport.py
搜索 self.banner_timeout 关键词，并将其参数改大即可，比如改为300s：
self.banner_timeout = 300
最后，重装paramiko即可。
3、下面的曲折、啰嗦的解决过程，不喜请跳过：
在谷歌搜到一个老外相关提问，虽然他说的是pysftp，其实也是基于paramiko：
https://stackoverflow.com/questions/34288526/pysft-paramiko-grequests-error-reading-ssh-protocol-banner/44493465#44493465
他最后给出了他的解决方案：
UPDATE:
It seems the problem is caused by importing the package grequests. If I do not import grequests, pysftp works as expected. The issue was raised before but has not been solved
意思是，在paramiko使用前，先import grequests，就能解决问题。我照做之后，发现对手头的现网环境无效，可能错误产生的原因不一样。
但是，我从老外的问题描述过程中，找到了解决方法，他是这样说的：
I have already tried changing the banner timeout from 15 seconds to 60 secs in the transport.py, but it did not solve the problem.
我看到有个timeout和transport.py，就想到现网那些报Error reading SSH protocol banner错误的机器也是非常卡，而且目测了下发起paramiko连接到报错的时间，基本是相同的。
于是系统中搜索，并找到了transport.py这个文件：
/usr/lib/python2.7/site-packages/paramiko/transport.py
并搜了下banner，发现果然有一个参数设置，而且和目测的超时基本一致！
于是，顺手修改成300S，并重新测试发现没任何效果，依然15S超时。接着打断点、甚至移走这个文件，问题依旧！！看来这个文件不会被引用。。。
回到最初的报错信息，发现里面显示的是：
build/bdist.linux-x86_64/egg/paramiko/transport.py
而系统里面搜不到这个问题，最后醍醐灌顶，发觉Python模块编译后，基本是以egg文件保存的，看来 必须修改源码才行了。
于是cd到paramiko的源码目录，执行搜索，找到2各transport.py文件：
[root@localhost:/data/software/paramiko-1.9]# find . -name transport.py
./paramiko/transport.py
./build/lib/paramiko/transport.py
尝试将文件中的 self.banner_timeout 值改成300，重新安装paramiko，结果一次性测试成功！
然后，我顺便在老外的帖子回答了下（请忽略蹩脚的英语），算是回馈吧！
二、paramiko远程执行后台脚本“阻塞”问题
我写的远程命令通道上线之后，发现在远程脚本中后台再执行另一个脚本，通道会一直等待后台脚本执行完成才会返回，有时甚至会僵死。
1、复现过程如下：
①、编写测试脚本
脚本1：test.sh
#!/bin/bash
sleep 30
echo test end
exit 0
脚本2：run.sh
#!/bin/bash
bash /tmp/test.sh &
echo run ok!
exit 0
脚本3：test.py
import paramiko
client = paramiko.SSHClient()
client = paramiko.SSHClient()
client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
client.connect(hostname='192.168.1.10', port=22, username='root', password='123456', timeout=300,allow_agent=False,look_for_keys=False)
stdin,stdout,stderr=client.exec_command("bash /tmp/run.sh")
result_info = ""
for line in stdout.readlines():
result_info += line
print result_info
将test.sh和run.sh传到远程服务器上，比如放到192.168.1.10:/tmp/下。
②、发起远程执行
在本地执行 python test.py，会发现整个脚本不会立即打印run ok，而是等30s之后才打印包括test.sh的所有输出信息。
2、解决办法
将远程脚本的标准输出stdout重定向到错误输出stderr即可，test.py 修改如下：
import paramiko
client = paramiko.SSHClient()
client = paramiko.SSHClient()
client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
client.connect(hostname='192.168.1.10', port=22, username='root', password='123456', timeout=300,allow_agent=False,look_for_keys=False)
stdin,stdout,stderr=client.exec_command("bash /tmp/run.sh 1>&2")
result_info = ""
for line in stderr.readlines():
result_info += line
print result_info
现在执行，就能立即得到结果了。其实原因很简单，因为stdout（标准输出），输出方式是行缓冲。输出的字符会先存放在缓冲区，等按下回车键时才进行实际的I/O操作，导致paramiko远程命令产生等待问题。而stderr（标准错误），是不带缓冲的，这使得出错信息可以直接尽快地显示出来。所以，这里只要将脚本执行的标准输出重定向到错误输出（1>&2），然后paramiko就可以使用stderr快速读取远程打屏信息了。
三、This operation would block forever 报错解决
这次扩容一个基于pramiko的自动化apiserver，结果发现在新环境执行远程命令或文件传输时，抛了如下报错：
2017-08-04 12:38:31,243 [ERROR] Exception: Error reading SSH protocol banner('This operation would block forever', <Hub at 0x38b02d0 epoll pending=0 ref=0 fileno=28>)
2017-08-04 12:38:31,244 [ERROR] Traceback (most recent call last):
2017-08-04 12:38:31,244 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/transport.py", line 1555, in run
2017-08-04 12:38:31,245 [ERROR] self._check_banner()
2017-08-04 12:38:31,245 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/transport.py", line 1681, in _check_banner
2017-08-04 12:38:31,245 [ERROR] raise SSHException('Error reading SSH protocol banner' + str(x))
2017-08-04 12:38:31,245 [ERROR] SSHException: Error reading SSH protocol banner('This operation would block forever', <Hub at 0x38b02d0 epoll pending=0 ref=0 fileno=28>)
2017-08-04 12:38:31,245 [ERROR]
2017-08-04 12:38:31,247 [INFO] Error reading SSH protocol banner('This operation would block forever', <Hub at 0x38b02d0 epoll pending=0 ref=0 fileno=28>)
总以为是python组件安装有问题，反反复复检查，最终发现居然是多装了一个插件导致的！
解决办法：
删除已经安装 greenlet插件即可，具体原因见后文:
rm -r /usr/local/
python2.7.5/lib/python2.7/site-packages/greenlet*
下面是"艰难险阻"的解决过程，不喜勿看：
1、看到报错，作为懒人第一时间就搜了下 【This operation would block forever', <Hub】这个关键词，发现没能get到解决方案。
2、按照经验，我先找到图中 _check_banner 函数如下：
def _check_banner(self):
# this is slow, but we only have to do it once
for i in range(100):
# give them 15 seconds for the first line, then just 2 seconds
# each additional line. (some sites have very high latency.)
if i == 0:
timeout = self.banner_timeout
else:
timeout = 2
buf = self.packetizer.readline(timeout)
except ProxyCommandFailure:
raise
except Exception, x:
raise SSHException('Error reading SSH protocol banner' + str(x))
if buf[:4] == 'SSH-':
break
self._log(DEBUG, 'Banner: ' + buf)
if buf[:4] != 'SSH-':
raise SSHException('Indecipherable protocol version "' + buf + '"')
# save this server version string for later
self.remote_version = buf
# pull off any attached comment
comment = ''
i = string.find(buf, ' ')
if i >= 0:
comment = buf[i+1:]
buf = buf[:i]
# parse out version string and make sure it matches
segs = buf.split('-', 2)
if len(segs) < 3:
raise SSHException('Invalid SSH banner')
version = segs[1]
client = segs[2]
if version != '1.99' and version != '2.0':
raise SSHException('Incompatible version (%s instead of 2.0)' % (version,))
self._log(INFO, 'Connected (version %s, client %s)' % (version, client))
3、很明显这个异常由 buf = self.packetizer.readline(timeout) 语句抛出，我印象中的粗暴定位方法就是不使用try，直接将此语句执行看看：
def _check_banner(self):
# this is slow, but we only have to do it once
for i in range(100):
# give them 15 seconds for the first line, then just 2 seconds
# each additional line. (some sites have very high latency.)
if i == 0:
timeout = self.banner_timeout
else:
timeout = 2
buf = self.packetizer.readline(timeout) # 我就加到，看看是从哪出来的异常
buf = self.packetizer.readline(timeout)
except ProxyCommandFailure:
raise
except Exception, x:
raise SSHException('Error reading SSH protocol banner' + str(x))
if buf[:4] == 'SSH-':
break
self._log(DEBUG, 'Banner: ' + buf)
.....
结果报错信息就更加具体了，如下所示：
2017-08-04 13:23:26,085 [ERROR] Unknown exception: ('This operation would block forever', <Hub at 0x20390f0 epoll pending=0 ref=0 fileno=27>)
2017-08-04 13:23:26,087 [ERROR] Traceback (most recent call last):
2017-08-04 13:23:26,088 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/transport.py", line 1555, in run
2017-08-04 13:23:26,088 [ERROR] self._check_banner()
2017-08-04 13:23:26,088 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/transport.py", line 1676, in _check_banner
2017-08-04 13:23:26,088 [ERROR] buf = self.packetizer.readline(timeout)
2017-08-04 13:23:26,088 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/packet.py", line 280, in readline
2017-08-04 13:23:26,088 [ERROR] buf += self._read_timeout(timeout)
2017-08-04 13:23:26,088 [ERROR] File "build/bdist.linux-x86_64/egg/paramiko/packet.py", line 468, in _read_timeout
2017-08-04 13:23:26,089 [ERROR] x = self.__socket.recv(128)
2017-08-04 13:23:26,089 [ERROR] File "/usr/local/python2.7.5/lib/python2.7/site-packages/gevent-1.1.2-py2.7-linux-x86_64.egg/gevent/_socket2.py", line 280, in recv
2017-08-04 13:23:26,089 [ERROR] self._wait(self._read_event)
2017-08-04 13:23:26,089 [ERROR] File "/usr/local/python2.7.5/lib/python2.7/site-packages/gevent-1.1.2-py2.7-linux-x86_64.egg/gevent/_socket2.py", line 179, in _wait
2017-08-04 13:23:26,089 [ERROR] self.hub.wait(watcher)
2017-08-04 13:23:26,089 [ERROR] File "/usr/local/python2.7.5/lib/python2.7/site-packages/gevent-1.1.2-py2.7-linux-x86_64.egg/gevent/hub.py", line 630, in wait
2017-08-04 13:23:26,089 [ERROR] result = waiter.get()
2017-08-04 13:23:26,089 [ERROR] File "/usr/local/python2.7.5/lib/python2.7/site-packages/gevent-1.1.2-py2.7-linux-x86_64.egg/gevent/hub.py", line 878, in get
2017-08-04 13:23:26,090 [ERROR] return self.hub.switch()
2017-08-04 13:23:26,090 [ERROR] File "/usr/local/python2.7.5/lib/python2.7/site-packages/gevent-1.1.2-py2.7-linux-x86_64.egg/gevent/hub.py", line 609, in switch
2017-08-04 13:23:26,090 [ERROR] return greenlet.switch(self)
2017-08-04 13:23:26,090 [ERROR] LoopExit: ('This operation would block forever', <Hub at 0x20390f0 epoll pending=0 ref=0 fileno=27>)
2017-08-04 13:23:26,090 [ERROR]
2017-08-04 13:23:26,093 [INFO] ('This operation would block forever', <Hub at 0x20390f0 epoll pending=0 ref=0 fileno=27>)
这次基本就定位到了gevent和greenlet这个真凶了！本以为是我的apiserver调用了gevent，结果定位了半天，确定并没有使用。而且印象中paramiko这个插件也没用到gevent，可这异常是怎么来的？
直到我再次在谷歌搜索【LoopExit: ('This operation would block forever', <Hub at】关键词，找到一个博客文章：http://www.hongquan.me/?p=178，总算知道是什么原因了！
具体原因：主要是因为 greenlet 里面有个run函数，覆盖了 paramiko 的transport.py 里面的同名函数，导致paramiko执行_check_banner时，实际调用了greenlet的run函数，因此报错！再次醉了！
《未完待续，更多问题，后续继续补充...》
