Skip to content

Commit

Permalink
[pfc_detect] fix RedisReply errors (#2040)
Browse files Browse the repository at this point in the history
**What I did**
It could be that queue counters are populated to FLEX COUNTER DB before the COUNTER_QUEUE_INDEX_MAP is populated which is in accordance to PortsOrch implementation. Fixing it by replacing the order - first the COUNTER_QUEUE_INDEX_MAP then the FLEX COUNTER DB might produce other issues if readers start reading the map and try to find the corresponding entry in COUNTERS DB. Such an order is also in alignment with other maps and counters. So seems like fixing it in the reader is more suitable fix.

Errors observed only for short period of time at system start when PFC WD is enabled:
```
Nov 10 15:12:19.001472 tgs-sonic-n2-s1 ERR syncd#SDK: :- guard: RedisReply catches system_error: command: *135#015#012$7#015#012EVALSHA#015#012$40#015#012b62081cc93943a4cbfec30de42638b435d31197c#015#012$3#015#012128#015#012$20#015#012oid:0x15000000000290#015#012$20#015#012oid:0x15000000000291#015#012$20#015#012oid:0x150000000002b1#015#012$20#015#012oid:0x150000000002b2#015#012$20#015#012oid:0x150000000002d2#015#012$20#015#012oid:0x150000000002d3#015#012$20#015#012oid:0x150000000002f3#015#012$20#015#012oid:0x150000000002f4#015#012$20#015#012oid:0x15000000000314#015#012$20#015#012oid:0x15000000000315#015#012$20#015#012oid:0x15000000000335#015#012$20#015#012oid:0x15000000000336#015#012$20#015#012oid:0x15000000000356#015#012$20#015#012oid:0x15000000000357#015#012$20#015#012oid:0x15000000000377#015#012$20#015#012oid:0x15000000000378#015#012$20#015#012oid:0x15000000000398#015#012$20#015#012oid:0x15000000000399#015#012$20#015#012oid:0x150000000003b9#015#012$20#015#012oid:0x150000000003ba#015#012$20#015#012oid:0x150000000003da#015#012$20#015#012oid:0x150000000003db#015#012$20#015#012oid:0x150000000003fb#015#012$20#015#012oid:0x150000000003fc#015#012$20#015#012oid:0x1500000000041c#015#012$20#015#012oid:0x1500000000041d#015#012$20#015#012oid:0x1500000000043d#015#012$20#015#012oid:0x1500000000043e#015#012$20#015#012oid:0x1500000000045e#015#012$20#015#012oid:0x1500000000045f#015#012$20#015#012oid:0x1500000000047f#015#012$20#015#012oid:0x15000000000480#015#012$20#015#012oid:0x150000000004a0#015#012$20#015#012oid:0x150000000004a1#015#012$20#015#012oid:0x150000000004c1#015#012$20#015#012oid:0x150000000004c2#015#012$20#015#012oid:0x150000000004e2#015#012$20#015#012oid:0x150000000004e3#015#012$20#015#012oid:0x15000000000503#015#012$20#015#012oid:0x15000000000504#015#012$20#015#012oid:0x15000000000524#015#012$20#015#012oid:0x15000000000525#015#012$20#015#012oid:0x15000000000545#015#012$20#015#012oid:0x15000000000546#015#012$20#015#012oid:0x15000000000566#015#012$20#015#012oid:0x15000000000567#015#012$20#015#012oid:0x15000000000587#015#012$20#015#012oid:0x15000000000588#015#012$20#015#012oid:0x150000000005a8#015#012$20#015#012oid:0x150000000005a9#015#012$20#015#012oid:0x150000000005c9#015#012$20#015#012oid:0x150000000005ca#015#012$20#015#012oid:0x150000000005ea#015#012$20#015#012oid:0x150000000005eb#015#012$20#015#012oid:0x1500000000060b#015#012$20#015#012oid:0x1500000000060c#015#012$20#015#012oid:0x1500000000062c#015#012$20#015#012oid:0x1500000000062d#015#012$20#015#012oid:0x1500000000064d#015#012$20#015#012oid:0x1500000000064e#015#012$20#015#012oid:0x1500000000066e#015#012$20#015#012oid:0x1500000000066f#015#012$20#015#012oid:0x1500000000068f#015#012$20#015#012oid:0x15000000000690#015#012$20#015#012oid:0x150000000006b0#015#012$20#015#012oid:0x150000000006b1#015#012$20#015#012oid:0x150000000006d1#015#012$20#015#012oid:0x150000000006d2#015#012$20#015#012oid:0x150000000006f2#015#012$20#015#012oid:0x150000000006f3#015#012$20#015#012oid:0x15000000000713#015#012$20#015#012oid:0x15000000000714#015#012$20#015#012oid:0x15000000000734#015#012$20#015#012oid:0x15000000000735#015#012$20#015#012oid:0x15000000000755#015#012$20#015#012oid:0x15000000000756#015#012$20#015#012oid:0x15000000000776#015#012$20#015#012oid:0x15000000000777#015#012$20#015#012oid:0x15000000000797#015#012$20#015#012oid:0x15000000000798#015#012$20#015#012oid:0x150000000007b8#015#012$20#015#012oid:0x150000000007b9#015#012$20#015#012oid:0x150000000007d9#015#012$20#015#012oid:0x150000000007da#015#012$20#015#012oid:0x150000000007fa#015#012$20#015#012oid:0x150000000007fb#015#012$20#015#012oid:0x1500000000081b#015#012$20#015#012oid:0x1500000000081c#015#012$20#015#012oid:0x1500000000083c#015#012$20#015#012oid:0x1500000000083d#015#012$20#015#012oid:0x1500000000085d#015#012$20#015#012oid:0x1500000000085e#015#012$20#015#012oid:0x1500000000087e#015#012$20#015#012oid:0x1500000000087f#015#012$20#015#012oid:0x1500000000089f#015#012$20#015#012oid:0x150000000008a0#015#012$20#015#012oid:0x150000000008c0#015#012$20#015#012oid:0x150000000008c1#015#012$20#015#012oid:0x150000000008e1#015#012$20#015#012oid:0x150000000008e2#015#012$20#015#012oid:0x15000000000902#015#012$20#015#012oid:0x15000000000903#015#012$20#015#012oid:0x15000000000923#015#012$20#015#012oid:0x15000000000924#015#012$20#015#012oid:0x15000000000944#015#012$20#015#012oid:0x15000000000945#015#012$20#015#012oid:0x15000000000965#015#012$20#015#012oid:0x15000000000966#015#012$20#015#012oid:0x15000000000986#015#012$20#015#012oid:0x15000000000987#015#012$20#015#012oid:0x150000000009a7#015#012$20#015#012oid:0x150000000009a8#015#012$20#015#012oid:0x150000000009c8#015#012$20#015#012oid:0x150000000009c9#015#012$20#015#012oid:0x150000000009e9#015#012$20#015#012oid:0x150000000009ea#015#012$20#015#012oid:0x15000000000a0a#015#012$20#015#012oid:0x15000000000a0b#015#012$20#015#012oid:0x15000000000a2b#015#012$20#015#012oid:0x15000000000a2c#015#012$20#015#012oid:0x15000000000a4c#015#012$20#015#012oid:0x15000000000a4d#015#012$20#015#012oid:0x15000000000a6d#015#012$20#015#012oid:0x15000000000a6e#015#012$20#015#012oid:0x15000000000a8e#015#012$20#015#012oid:0x15000000000a8f#015#012$20#015#012oid:0x15000000000aaf#015#012$20#015#012oid:0x15000000000ab0#015#012$1#015#0122#015#012$8#015#012COUNTERS#015#012$3#015#012100#015#012$2#015#012''#15#012, reason: ERR Error running script (call to f_b62081cc93943a4cbfec30de42638b435d31197c): @user_script:39: user_script:39: attempt to concatenate local 'queue_index' (a boolean value) : Input/output error

```

**Why I did it**

If queue_index or port_id is not defined don't run the rest of the LUA script logic.

**How I verified it**

Running it on the switch and verify no errors.
  • Loading branch information
stepanblyschak authored Dec 6, 2021
1 parent b38f527 commit 8123437
Show file tree
Hide file tree
Showing 6 changed files with 305 additions and 275 deletions.
101 changes: 53 additions & 48 deletions orchagent/pfc_detect_barefoot.lua
Original file line number Diff line number Diff line change
Expand Up @@ -36,63 +36,68 @@ for i = n, 1, -1 do

local queue_index = redis.call('HGET', 'COUNTERS_QUEUE_INDEX_MAP', KEYS[i])
local port_id = redis.call('HGET', 'COUNTERS_QUEUE_PORT_MAP', KEYS[i])
local pfc_rx_pkt_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PKTS'
local pfc_duration_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PAUSE_DURATION'
-- If there is no entry in COUNTERS_QUEUE_INDEX_MAP or COUNTERS_QUEUE_PORT_MAP then
-- it means KEYS[i] queue is inserted into FLEX COUNTER DB but the corresponding
-- maps haven't been updated yet.
if queue_index and port_id then
local pfc_rx_pkt_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PKTS'
local pfc_duration_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PAUSE_DURATION'

-- Get all counters
local occupancy_bytes = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES')
local packets = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS')
local pfc_rx_packets = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key)
local pfc_duration = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_duration_key)
-- Get all counters
local occupancy_bytes = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES')
local packets = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS')
local pfc_rx_packets = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key)
local pfc_duration = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_duration_key)

if occupancy_bytes and packets and pfc_rx_packets and pfc_duration then
occupancy_bytes = tonumber(occupancy_bytes)
packets = tonumber(packets)
pfc_rx_packets = tonumber(pfc_rx_packets)
pfc_duration = tonumber(pfc_duration)
if occupancy_bytes and packets and pfc_rx_packets and pfc_duration then
occupancy_bytes = tonumber(occupancy_bytes)
packets = tonumber(packets)
pfc_rx_packets = tonumber(pfc_rx_packets)
pfc_duration = tonumber(pfc_duration)

local packets_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last')
local pfc_rx_packets_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
local pfc_duration_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last')
-- DEBUG CODE START. Uncomment to enable
local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
-- DEBUG CODE END.
local packets_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last')
local pfc_rx_packets_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
local pfc_duration_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last')
-- DEBUG CODE START. Uncomment to enable
local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
-- DEBUG CODE END.

-- If this is not a first run, then we have last values available
if packets_last and pfc_rx_packets_last and pfc_duration_last then
packets_last = tonumber(packets_last)
pfc_rx_packets_last = tonumber(pfc_rx_packets_last)
pfc_duration_last = tonumber(pfc_duration_last)
-- If this is not a first run, then we have last values available
if packets_last and pfc_rx_packets_last and pfc_duration_last then
packets_last = tonumber(packets_last)
pfc_rx_packets_last = tonumber(pfc_rx_packets_last)
pfc_duration_last = tonumber(pfc_duration_last)

-- Check actual condition of queue being in PFC storm
if (occupancy_bytes > 0 and packets - packets_last == 0 and pfc_rx_packets - pfc_rx_packets_last > 0) or
-- DEBUG CODE START. Uncomment to enable
(debug_storm == "enabled") or
-- DEBUG CODE END.
(occupancy_bytes == 0 and packets - packets_last == 0 and (pfc_duration - pfc_duration_last) > poll_time * 0.8) then
if time_left <= poll_time then
redis.call('HDEL', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
redis.call('HDEL', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last')
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","storm"]')
is_deadlock = true
time_left = detection_time
-- Check actual condition of queue being in PFC storm
if (occupancy_bytes > 0 and packets - packets_last == 0 and pfc_rx_packets - pfc_rx_packets_last > 0) or
-- DEBUG CODE START. Uncomment to enable
(debug_storm == "enabled") or
-- DEBUG CODE END.
(occupancy_bytes == 0 and packets - packets_last == 0 and (pfc_duration - pfc_duration_last) > poll_time * 0.8) then
if time_left <= poll_time then
redis.call('HDEL', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
redis.call('HDEL', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last')
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","storm"]')
is_deadlock = true
time_left = detection_time
else
time_left = time_left - poll_time
end
else
time_left = time_left - poll_time
end
else
if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","restore"]')
if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","restore"]')
end
time_left = detection_time
end
time_left = detection_time
end
end

-- Save values for next run
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last', packets)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
if is_deadlock == false then
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last', pfc_rx_packets)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last', pfc_duration)
-- Save values for next run
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last', packets)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
if is_deadlock == false then
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last', pfc_rx_packets)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_duration_key .. '_last', pfc_duration)
end
end
end
end
Expand Down
95 changes: 50 additions & 45 deletions orchagent/pfc_detect_broadcom.lua
Original file line number Diff line number Diff line change
Expand Up @@ -35,61 +35,66 @@ for i = n, 1, -1 do

local queue_index = redis.call('HGET', 'COUNTERS_QUEUE_INDEX_MAP', KEYS[i])
local port_id = redis.call('HGET', 'COUNTERS_QUEUE_PORT_MAP', KEYS[i])
local pfc_rx_pkt_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PKTS'
local pfc_on2off_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_ON2OFF_RX_PKTS'
-- If there is no entry in COUNTERS_QUEUE_INDEX_MAP or COUNTERS_QUEUE_PORT_MAP then
-- it means KEYS[i] queue is inserted into FLEX COUNTER DB but the corresponding
-- maps haven't been updated yet.
if queue_index and port_id then
local pfc_rx_pkt_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_RX_PKTS'
local pfc_on2off_key = 'SAI_PORT_STAT_PFC_' .. queue_index .. '_ON2OFF_RX_PKTS'

-- Get all counters
local occupancy_bytes = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES')
local packets = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS')
local pfc_rx_packets = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key)
local pfc_on2off = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key)
local queue_pause_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS')
-- Get all counters
local occupancy_bytes = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_CURR_OCCUPANCY_BYTES')
local packets = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS')
local pfc_rx_packets = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key)
local pfc_on2off = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key)
local queue_pause_status = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS')

if occupancy_bytes and packets and pfc_rx_packets and pfc_on2off and queue_pause_status then
occupancy_bytes = tonumber(occupancy_bytes)
packets = tonumber(packets)
pfc_rx_packets = tonumber(pfc_rx_packets)
pfc_on2off = tonumber(pfc_on2off)
if occupancy_bytes and packets and pfc_rx_packets and pfc_on2off and queue_pause_status then
occupancy_bytes = tonumber(occupancy_bytes)
packets = tonumber(packets)
pfc_rx_packets = tonumber(pfc_rx_packets)
pfc_on2off = tonumber(pfc_on2off)

local packets_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last')
local pfc_rx_packets_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
local pfc_on2off_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last')
local queue_pause_status_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last')
local packets_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last')
local pfc_rx_packets_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last')
local pfc_on2off_last = redis.call('HGET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last')
local queue_pause_status_last = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last')

-- DEBUG CODE START. Uncomment to enable
local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
-- DEBUG CODE END.
-- DEBUG CODE START. Uncomment to enable
local debug_storm = redis.call('HGET', counters_table_name .. ':' .. KEYS[i], 'DEBUG_STORM')
-- DEBUG CODE END.

-- If this is not a first run, then we have last values available
if packets_last and pfc_rx_packets_last and pfc_on2off_last and queue_pause_status_last then
packets_last = tonumber(packets_last)
pfc_rx_packets_last = tonumber(pfc_rx_packets_last)
pfc_on2off_last = tonumber(pfc_on2off_last)
-- If this is not a first run, then we have last values available
if packets_last and pfc_rx_packets_last and pfc_on2off_last and queue_pause_status_last then
packets_last = tonumber(packets_last)
pfc_rx_packets_last = tonumber(pfc_rx_packets_last)
pfc_on2off_last = tonumber(pfc_on2off_last)

-- Check actual condition of queue being in PFC storm
if (pfc_rx_packets - pfc_rx_packets_last > 0 and pfc_on2off - pfc_on2off_last == 0 and queue_pause_status_last == 'true' and queue_pause_status == 'true') or
(debug_storm == "enabled") then
if time_left <= poll_time then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","storm"]')
is_deadlock = true
time_left = detection_time
-- Check actual condition of queue being in PFC storm
if (pfc_rx_packets - pfc_rx_packets_last > 0 and pfc_on2off - pfc_on2off_last == 0 and queue_pause_status_last == 'true' and queue_pause_status == 'true') or
(debug_storm == "enabled") then
if time_left <= poll_time then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","storm"]')
is_deadlock = true
time_left = detection_time
else
time_left = time_left - poll_time
end
else
time_left = time_left - poll_time
end
else
if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","restore"]')
if pfc_wd_action == 'alert' and pfc_wd_status ~= 'operational' then
redis.call('PUBLISH', 'PFC_WD_ACTION', '["' .. KEYS[i] .. '","restore"]')
end
time_left = detection_time
end
time_left = detection_time
end
end

-- Save values for next run
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last', queue_pause_status)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last', packets)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last', pfc_rx_packets)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last', pfc_on2off)
-- Save values for next run
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_ATTR_PAUSE_STATUS_last', queue_pause_status)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'SAI_QUEUE_STAT_PACKETS_last', packets)
redis.call('HSET', counters_table_name .. ':' .. KEYS[i], 'PFC_WD_DETECTION_TIME_LEFT', time_left)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_rx_pkt_key .. '_last', pfc_rx_packets)
redis.call('HSET', counters_table_name .. ':' .. port_id, pfc_on2off_key .. '_last', pfc_on2off)
end
end
end
end
Expand Down
Loading

0 comments on commit 8123437

Please sign in to comment.