Description

On QFX5K platforms with EVPN-VXLAN, the dcpfe process may crash when the size of the Local Bias Filter Bitmap string exceeds 256 characters.

Symptoms

On QFX5K platforms with EVPN-VXLAN and ESI, a Local Bias Filter is installed to avoid sending the BUM Traffic (Broadcast, Unknown Unicast, and Multicast) back to the source. Each of the ESI ports local to the Leaf is added to Port Bitmap to drop the traffic going back to the source if the BUM traffic is originated from its peer leaf ESI. The Local Bias Filter Bitmap is added when the ESI is configured regardless of whether the port is up or down as long as optics is inserted.

Once the Bitmap string reaches the 256 max limit, it may lead to a PFE crash. The PFE crash will continue until the size of the string is reduced to less than 256 characters.

Taking the example of QFX5100-96S where there are 96 10G ports configured in a contiguous fashion without any gap between ESI ports, the string would be "xe0-xe95" which is only 8 characters. However, taking the worst-case scenario, after every 2 ESI contiguous ports, the 3rd port is non-ESI or unused, then the string will be as shown below, which is 313 characters and will lead to a PFE crash:

"xe0-xe1,xe3-xe4,xe6-xe7,xe9-xe10,xe12-xe13,xe15-xe16,xe18-xe19,xe21-xe22,xe24-xe25,xe27-xe28,xe30-xe31,xe33-xe34,xe36-xe37,xe39-xe40,xe42-xe43,xe45-xe46,xe48-xe49,xe51-xe52,xe54-xe55,xe57-xe58,xe60-xe61,xe63-xe64,xe66-xe67,xe69-xe70,xe72-xe73,xe75-xe76,xe78-xe79,xe81-xe82,xe84-xe85,xe87-xe88,xe90-xe91,xe93-xe94".

The string length can be validated with the following command:

root@device:0% cprod -A fpc0 -c "show shim virtual evpn_filter"
EVPN is-config flag is TRUE

EVPN Default Filter Info:
EVPN Default Filter ID: 802
EVPN Higig Default Filter ID: 803
EVPN Default Filter Ref Count: 337
EVPN MCAST Default Filter ID: 804
EVPN MCAST Default Ref Cnount: 337

EVPN Local Bias Filter Info:

EVPN SVP GPORT 0x80001400
Filter ID: 813
Higig Filter ID: 814
EVPN IGMP Filter ID: 815
The Ports in the Bitmap are xe10-xe11,xe13-xe14,xe16-xe17,xe19-xe20,xe22-xe23,xe25-xe26,xe28-xe29,xe31-xe32,xe34-xe35,xe37-xe38,xe40-xe41,xe43-xe44,xe46-xe47,xe49-xe50,xe52-xe53,xe55-xe56,xe58-xe59,xe62-xe63,xe65-xe68,xe70-xe71,xe73-xe74,xe76-xe77,xe79-xe80,xe82-xe83,xe85-xe86,xe88

The core-dump files can be seen by executing the CLI command "show system core-dumps".

user@device> show system core-dumps no-forwarding
-rw-r--r--  1 root  wheel   12904585 Date  /var/tmp/dcpfe0.dcpfe.23779.1610460688.core.tgz
-rw-r--r--  1 root  wheel   16131498 Date  /var/tmp/dcpfe0.dcpfe.2589.1610440615.core.tgz
-rw-r--r--  1 root  wheel   12812609 Date  /var/tmp/dcpfe0.dcpfe.5986.1610440809.core.tgz
total files: 3

Solution

It is recommended to avoid a gap between ESI ports so the Local Bias Port Bitmap string size does not exceed 256 characters.

Related Information