A little more detail:
- It appears to me that the ODL OVSDB subsystem sends an initial query to the switch when an OVSDB connection comes up
- If the response to this query is 'too large' (current 100,000 bytes), ODL closes the connection.
- If the switch is configured to open the OVSDB connection, this process repeats indefinitely and the OVSDB subsystem isn't usable (at least for that switch)
A lot of detail:
Running code that I built myself from the stable/helium branch (more or less at stable helium 2).
I had one switch in my system with A LOT of ovsdb rows and I hit this problem
- My switch is configured to connect to my controller when the controller starts up.
- My controller starts and the switch connects
- The OVSDB subsystem begins pulling information from my switch
- My controller closes the OVSDB connection
- My switch attempts to reconnect
- repeat forever
I eventually was able to set a break point and get a stack trace. The problem is in
ovsdb/library/src/main/java/org/opendaylight/ovsdb/lib/jsonrpc/JsonRpcDecoder.java around line 116 (in the decode(ChannelHandlerContext ctx, ByteBuf buf, List<Object> out) method)
for (; i < buf.writerIndex(); i++) {
switch (buf.getByte
) {
case '
':
if (!inS) rightCurlies++;
break;
case '"':
}
if (leftCurlies != 0 && leftCurlies == rightCurlies && !inS)
{ ByteBuf slice = buf.readSlice(1 + i - buf.readerIndex()); JsonParser jp = jacksonJsonFactory.createParser(new ByteBufInputStream(slice)); JsonNode root = jp.readValueAsTree(); out.add(root); leftCurlies = rightCurlies = lastRecordBytes = 0; recordsRead++; break; }if (i - buf.readerIndex() >= maxFrameLength)
{ fail(ctx, i - buf.readerIndex()); }...
private void fail(ChannelHandlerContext ctx, long frameLength) {
logger.error("JSON too big. JSON content exceeded limit of {} bytes", maxFrameLength);
if (frameLength > 0)
else
{ ctx.fireExceptionCaught( new TooLongFrameException( "frame length exceeds " + maxFrameLength + " - discarding")); }}