TCP support #226
TCP support #226
Conversation
| from ..dgram.base import DgramSocketTransport | ||
|
|
||
| # Ignore these socket errors | ||
| sockErrors = {errno.ESHUTDOWN: True, |
There was a problem hiding this comment.
Since we are reusing this map across tcp/udp transports, would it make sense to move it to some common place e.g. asyncore.base module or a common base class?
| sockErrors[errno.EBADFD] = True | ||
|
|
||
|
|
||
| class StreamSocketTransport(DgramSocketTransport): |
There was a problem hiding this comment.
This whole class seems to be largely a copy of DgramSocketTransport. Could we make this common code shared between stream and dgram classes? May be by pushing it to AbstractSocketTransport?
| if header_data: | ||
| #1st byte is 0x30 - sequence | ||
| seq_len = header_data[1] | ||
| if seq_len & 0x80 != 0: # BER long definite length |
There was a problem hiding this comment.
Finding start of message in the stream is hard indeed...
It would be cleaner if we keep transport code independent from the data being transferred. That would involve a bit of refactoring in the protocol engine, though.
Here is where the read message goes from transport/networking code into the protocol engine. I am thinking if we make receiveMessage returning unused bytes (as returned by the underlying pyasn1 decoder) and "unread" them back into some internal queue like __inQueue, but (in case of stream transport) it should probably just a (growing/shrinking) string rather than a list...? Just like you have it in remaining_bytes.
That would let us remove this header finding heuristics completely. And, most importantly, make searching way more reliable.
One of the interesting problems along this way would be to detect and eventually reset the growing inQueue if decoder repeatedly fails to decode it. May be this could be based on a reasonable SNMP message size, like if the buffer grows beyond 4K - drop it.
Does it make sense?
| @@ -19,6 +20,7 @@ | |||
| # Transports | |||
| snmpUDPDomain = udp.snmpUDPDomain | |||
| snmpUDP6Domain = udp6.snmpUDP6Domain | |||
| snmpTCPDomain = tcp.snmpTCPDomain | |||
There was a problem hiding this comment.
Just noting that supporting TCP-over-IP6 would be comprise a great follow-up patch! ;-)
| Datastore (LCD) managed by :py:class:`~pysnmp.hlapi.SnmpEngine` | ||
| class instance. | ||
|
|
||
| See :RFC:`1906#section-3` for more information on the UDP transport mapping. |
|
|
||
| Examples | ||
| -------- | ||
| >>> from pysnmp.hlapi.v3arch.asyncore import UdpTransportTarget |
| @@ -36,7 +36,7 @@ | |||
|
|
|||
| def inet_pton(address_family, ip_string): | |||
| if address_family == socket.AF_INET: | |||
| return inet_aton(ip_string) | |||
| return socket.inet_aton(ip_string) | |||
There was a problem hiding this comment.
Ah, seems like an unrelated bug! Let me fix this in master...
There was a problem hiding this comment.
Just a quick note that this problem has been fixed in master.
|
What is the status of adding the TCP support ? Is there an Alpha release for this feature ? |
Added TCP support based on asyncore. Tested on Windows on python versions 2.7 and 3.5 connecting to a SNMPv3 target (with authPriv) and Travis tests passing.
This can be a good point to start and solve #56.