EIGRP Neighbor States and DUAL Explained: Passive, Active, and Stuck in Active
EIGRP does not have a long ladder of neighbour states like OSPF or BGP. Its important states belong to routes, not neighbours: a route is Passive when it is settled, and Active when its router has lost the best path, has no proved backup, and is asking every neighbour for help. This guide explains that split in plain English — successor, feasible successor, and the one inequality that proves a backup is loop-free — then walks one loss with a backup and one without, and explains stuck in Active and the neighbour-change log line engineers actually meet.
The short answer
What are EIGRP's states? Unlike OSPF and BGP, EIGRP does not march neighbours through a long state ladder. A neighbour is simply formed after Hellos and an initial exchange, and then it is up — or it is down, with a reason logged. The states engineers must understand belong to each route inside DUAL, EIGRP's calculation engine: Passive, meaning stable, and Active, meaning the router is actively asking its neighbours for a new path. Confusing those two levels — neighbour up/down versus route Passive/Active — is the most common EIGRP misunderstanding.
Think of a delivery driver with a regular route. Passive is an ordinary day: the best road is known, and a backup road has already been checked and written on the card in the cab. If the main road closes and the backup is on the card, the driver just takes it — still an ordinary day. Active is the day the main road closes and there is no checked backup: now the driver has to phone every depot and wait for every answer before choosing. Stuck in Active is still waiting on the phone long after everyone should have answered.
First level: the neighbour — up, or down with a reason
EIGRP neighbours find each other with Hello packets sent to the multicast address 224.0.0.10 on EIGRP-enabled interfaces. On most links Hellos go every 5 seconds and the hold time — how long a router waits without hearing any EIGRP packet before declaring the neighbour down — is 15 seconds by default. For a neighbour relationship to form, the routers must be in the same EIGRP autonomous system, pass any configured authentication, and agree on the K values, the weights EIGRP uses in its metric calculation. Hello and hold timers themselves do not have to match, because each router tells the other its hold time in its Hellos.
Once formed, there is no ExStart-style ladder to watch. The useful evidence is in show ip eigrp neighbors: Hold counts down the seconds left before the neighbour is declared down; Uptime shows how long the relationship has stood — a neighbour whose uptime is always small is flapping, whatever the log says; SRTT is the smoothed round-trip time, in milliseconds, for an EIGRP packet to that neighbour and back as an acknowledgment; Q Cnt is how many EIGRP packets are queued waiting to be sent to that neighbour, and it should normally be 0 — a queue that stays above zero means packets are not getting through or not being acknowledged; and Seq Num is the sequence number of the last reliable packet received from that neighbour.
Every formation and loss is logged as %DUAL-5-NBRCHANGE when neighbour-change logging is on (it is on by default on modern Cisco IOS). The words after the neighbour's address are the evidence. Read them before you change anything, using the decoder entry for %DUAL-5-NBRCHANGE, which covers each reason in full.
| Log ending | What it proves | What to check first |
|---|---|---|
| is up: new adjacency | A neighbour relationship has just formed and the initial update exchange has happened. | Nothing, if it follows a change or a start-up. Everything, if it repeats every few minutes with a matching down message. |
| is down: holding time expired | No EIGRP packet — Hello or any other — arrived from that neighbour during the whole hold-time interval. | The link itself: errors, drops, congestion, or a one-way path that passes traffic in only one direction. |
| is down: retry limit exceeded | A reliable EIGRP packet (an Update, Query or Reply that needs an acknowledgment) was retransmitted without being acknowledged until the retry limit ran out. | Delivery on that link — a growing Q Cnt or rising SRTT in the neighbour table points the same way. |
| is down: peer restarted / resync | The neighbour itself signalled a restart or a resynchronisation, rather than simply going silent. | What happened on the far router at that time: a reload, a process restart, or a configuration change. |
Second level: the route — Passive and Active in DUAL
DUAL keeps every route in the topology table, and each route is in one of two states. Passive (P) means the route is stable: the router is not running a calculation for it, and the successor in the routing table can be trusted. Passive is the normal state for essentially every route, essentially all of the time. Active (A) means the route's successor has been lost, no feasible successor was available, and the router has started a diffusing computation — it has sent Query packets to its neighbours and cannot finish choosing a new path until every neighbour has sent a Reply. You can watch both with show ip eigrp topology, which prefixes each entry P or A, and list only the unfinished ones with show ip eigrp topology active.
The fork that decides an EIGRP failure: a proved backup keeps the route Passive; no backup sends it Active, where it must wait for every Reply.
The one inequality: successor, feasible successor, and RD < FD
For each destination, DUAL compares two numbers. Your feasible distance (FD) is your own best total cost to the destination. Each neighbour advertises its own cost, its reported distance (RD). The neighbour with the lowest resulting total cost becomes the successor. A different neighbour becomes a feasible successor — a backup DUAL trusts without asking anyone — only if it passes the feasibility condition:
Reported distance of the backup neighbour < your feasible distance (RD < FD)
The logic is simple enough to say in plain English: if that neighbour is already closer to the destination than you have ever been, its path cannot be looping back through you — a loop through you would have to be longer than your own best distance, not shorter. RFC 7868, section 3.3, states the honest limit too: every path that meets the feasibility condition is guaranteed loop-free, but not every loop-free path meets it. DUAL would rather reject a backup that might be fine than accept one it cannot prove safe.
A worked example, with this article's own round numbers for illustration — not a measurement from any real network. Your router reaches 10.4.0.0/24 through neighbour A with a total cost of 33280, so FD is 33280 and A is the successor. Neighbour B advertises an RD of 30720, which is less than 33280, so B passes the condition and waits as feasible successor. Neighbour C advertises an RD of 40960, which is not less than 33280, so C may hold a perfectly usable path in reality — but DUAL cannot prove it loop-free, so C is not a feasible successor and will not be used as an instant backup.
Two losses, end to end
Loss with a backup. Neighbour A fails. The route to 10.4.0.0/24 loses its successor, but B is already in the topology table as a feasible successor. DUAL promotes B to successor immediately, the routing table is updated, and the route never leaves Passive. No Queries are sent and no neighbour is waited on. This instant, silent repair is the whole point of pre-proving backups, and it is why two EIGRP networks can react so differently to the same cable pull.
Loss with no backup. Now suppose only A and C existed. A fails, C does not pass the feasibility condition, so there is no feasible successor. The route goes Active. Your router sends Queries to its remaining neighbours asking, in effect, what is your distance to this destination now? A neighbour that still has a working path replies with it. A neighbour whose own path ran through your router cannot answer at once — it must run its own DUAL calculation, and possibly send its own Queries further out, before it can reply. Your route stays Active until the last Reply arrives; only then does DUAL pick the best replying path, install a new successor, and return the route to Passive. Queries therefore spread exactly as far as the loss matters — which is why design, below, is about limiting that spread.
Stuck in Active: when a Reply never comes
Active is meant to last moments. If a router has not received a Reply to a Query from one or more neighbours within the time allowed — approximately 3 minutes by default, the Active timer — Cisco's guidance says the route is stuck in Active (SIA): the router logs %DUAL-3-SIA, clears the neighbour that did not reply, and removes the routes beyond it so everything can reconverge. Modern IOS softens the cliff edge with SIA-Query and SIA-Reply messages sent at the halfway point, about 90 seconds, asking a silent neighbour whether it is still working on the question — a neighbour that answers an SIA-Query is still alive and waiting on its own Replies, so it is given more time instead of being cleared on suspicion alone.
An SIA is never the root cause; it is a receipt proving a Query or a Reply was lost, delayed, or could not be generated. Work it in that order: is the unresponsive neighbour itself healthy, or overloaded? Is the link to it dropping or congesting reliable EIGRP packets — check the neighbour table for a Q Cnt stuck above zero and a rising SRTT, which say packets are queueing instead of being acknowledged. And is the network designed so one loss asks half the company? Two controls bound the blast radius: route summarisation, because a router that only knows a summary does not need to be queried for a detail inside it, and EIGRP stub configuration on edge routers, because neighbours do not send Queries to a stub at all — a stub is told routes, it is never asked to find them. A network that sees repeated SIAs has a delivery problem or a query-scope problem, and usually the logs tell you which: the same neighbour failing to reply points at delivery; many neighbours, many routes, one distant loss points at scope.
When a neighbour change is normal — and when it is a flap
Normal: one 'is up: new adjacency' after a router starts, an interface is enabled, or a deliberate change is made; a matching down message with a peer-restart reason while the far end was being worked on. The event is single, explained, and the uptime then grows for days.
A flap: the same neighbour cycling down and up, especially 'holding time expired' repeating. Each cycle is not free: every route that used that neighbour must be re-evaluated, and any route with no feasible successor goes Active and queries outward, so one bad link can keep a whole area recalculating. Read the reason word first, then the link counters on both ends, then the neighbour table's Q Cnt and SRTT — in that order — before touching EIGRP timers. Raising the hold time on a dropping link does not repair the link; it only makes EIGRP notice real failures more slowly.
For the companion state machines, read OSPF neighbor states explained — a true ladder, where the stuck state names the failed proof — and BGP neighbor states explained, where the first three states are all about building the TCP connection. EIGRP's lesson is different: watch the route's state and the neighbour's reason word, not a ladder.
Four mistakes that waste the evidence
1. Looking for an EIGRP neighbour ladder that does not exist. Time spent hunting for an EIGRP equivalent of ExStart is time not spent on the two things that carry the evidence: the route's P or A in the topology table, and the reason word in the NBRCHANGE log.
2. Calling every Active route a fault. Active for seconds during a real topology change is DUAL working as designed. Active that persists, repeats for the same prefix, or ends in %DUAL-3-SIA is the fault.
3. Assuming the second-best path is the backup. Only a neighbour with RD < FD is a feasible successor. The second-lowest cost alone proves nothing about loops, and DUAL will not promote such a neighbour instantly — the route will go Active instead.
4. Clearing the neighbour before reading the table. A clear throws away the uptime, Q Cnt, SRTT and the Active entries that show which Reply was missing. Record show ip eigrp neighbors and show ip eigrp topology active first, then change one thing.
Try it yourself
Run a two-router EIGRP configuration pair through the free Network Config Verifier and look for the mismatches — autonomous system, K values, network statements — that stop a neighbour forming at all. Then keep the %DUAL-5-NBRCHANGE decoder entry open next time the log moves: the reason word tells you whether you are looking at silence, at unacknowledged delivery, or at a neighbour that restarted itself.
Fact-check sources: route states Passive and Active, the feasibility condition as the source node condition (a neighbour's reported distance less than this router's feasible distance), successor and feasible successor definitions, and FD as the least-known distance since the last Active-to-Passive transition — RFC 7868, sections 3.2, 3.3 and 5. Stuck in Active meaning no Reply within about 3 minutes, the neighbour being cleared, and %DUAL-3-SIA — Cisco's 'What Does the EIGRP DUAL-3-SIA Error Message Mean?' document. NBRCHANGE reasons (holding time expired when no EIGRP packet arrived during the hold time; retry limit exceeded when a reliable packet was not acknowledged; new adjacency on an initial update) and neighbour-table fields (Hold, Uptime, SRTT, RTO, Q Cnt, Seq Num) — Cisco's 'Troubleshoot EIGRP Common Issues' and IOS command references. The 10.4.0.0/24 costs (33280, 30720, 40960) are this article's own illustrative numbers, chosen so the RD < FD test can be checked by eye; they are not measurements.
Related guides
- In EIGRP the states that matter are route states in DUAL: Passive means the route is stable and not being recalculated; Active means the router has lost its successor, has no feasible successor, and is running a diffusing computation — sending Queries and waiting for Replies from neighbours before it can settle again.
- The successor is the neighbour on the lowest-cost path, and its route goes in the routing table. A feasible successor is a pre-proved backup: it qualifies only when its reported distance is strictly less than this router's feasible distance (RD < FD). That inequality is the feasibility condition in RFC 7868, and it guarantees the backup does not loop back through this router.
- If a feasible successor exists when the main path fails, the router promotes it and the route stays Passive — no Queries, no waiting. If none exists, the route goes Active. If Replies do not all arrive within about 3 minutes, the route is stuck in Active (SIA), the unresponsive neighbour is cleared, and %DUAL-3-SIA is logged — summarisation and stub routers limit how far Queries spread, which is the main defence.
- %DUAL-5-NBRCHANGE is a neighbour up/down log, not a route state: 'is up: new adjacency' is a normal formation, while 'is down: holding time expired' means no EIGRP packet arrived within the hold time (15 seconds by default on most links, with Hellos every 5 seconds) and 'retry limit exceeded' means a reliable EIGRP packet was not acknowledged. A single up after a config change is normal; the same pair flapping is a link or delivery problem.